JPMorgan's Cembalest Warns 2031 US Fiscal Crossover Threatens Markets

摩根大通Cembalest警告：2031年美国财政交叉点威胁市场 到2031年，不断上升的强制性支出和利息成本将消耗美国联邦税收收入的100%，从而迫使债务评级下调和财政结构性改革。 摩根资产管理（J.P. Morgan Asset Management）的Michael Cembalest与主持人Josh Brown和Michael Batnick共同探讨，评估了美国经济的韧性及其长期结构性风险。讨论驳斥了围绕美元主导地位、央行购金以及外国减持美债的常见看空论调，同时指出联邦债务将在2031至2032年左右迎来关键的交叉点。此外，与会嘉宾还分析了复杂的人工智能格局，探讨了企业高昂的Token成本、软件颠覆、定制半导体竞争，以及AI发现软件漏洞所引发的日益严峻的网络安全脆弱性。 相比昂贵的美国前沿实验室模型，企业偏好正转向微调后的开放权重AI模型和低成本的中国开源方案。 (0:00) Michael Cembalest自视为一名“分析型爱国者”，并指出强劲的美国净移民倾向和持续创新是经济活力的关键证明。 (13:06) 在过去15到18年里，与全球基准相比，看空美国经济和美股几乎是一个彻底落败的交易。 (16:32) 尽管唱衰论调不断，美元在国际支付额、外汇储备和跨境贸易指标中依然保持着压倒性的主导地位。 (19:28) 央行黄金配置在外汇储备中占比上升，主要是由于按市值计价的价格上涨，而非大规模的实物黄金增持。 (23:56)

J.P. Morgan Asset Management’s Michael Cembalest argues that while the structural dominance of the U.S. economy and the dollar remains firmly entrenched, investors face escalating medium-term risks from an impending 2031–2032 federal debt crossover, debt-fueled AI hyperscaler spending, and asymmetric cybersecurity vulnerabilities. Structural Dominance of U.S. Equities and the Dollar (16:32) Cembalest observes that an aggressive allocation toward U.S. equities, credit, and the dollar has consistently outperformed international alternatives since the 2008 financial crisis. High price-to-earnings valuations on U.S. equities are largely matched by superior corporate earnings growth rates, aligning price-to-earnings-to-growth (PEG) ratios with global benchmarks. Concurrently, Cembalest dismisses near-term 'dollar bear' theories, noting that while the U.S. produces approximately 20% of global GDP, its currency represents roughly double that share across SWIFT payments, foreign exchange transactions, and trade invoicing. Purported alternatives face major structural roadblocks. Although China remains an economic powerhouse, Cembalest emphasizes that its closed capital accounts, excessive money supply growth, and rigid controls trigger an estimated $800 billion to $1 trillion in annual capital flight. Research by Brad Setser shows China continues to hold U.S. debt by routing purchases through custodial accounts in Belgium rather than executing a buyer's strike. Furthermore, Cembalest clarifies that surging central bank gold reserves primarily reflect mark-to-market price appreciation rather than an increase in allocated physical troy ounces. The 2031 Fiscal Crossover and Growth Bottlenecks (32:44) Cembalest outlines a mathematical timeline for the federal deficit, warning of a 'crossover point' by 2031 or 2032 when 100% of federal tax revenues will be consumed exclusively by mandatory entitlement spending and debt interest. He projects that credit rating agencies may issue formal warnings or downgrades two to three years earlier, around 2028 or 2029, to force legislative entitlement reform. While the U.S. previously reduced its debt-to-GDP ratio from 100% to 60% under President Eisenhower, Cembalest notes that was driven by an extraordinary post-WWII economic expansion that cannot easily be replicated under current demographic constraints. Near-term economic growth is further restrained by policy friction. Deregulation structurally lags corporate implementation by three to five years, while tariffs create critical supply-chain imbalances. For example, while 80% of direct semiconductor imports benefit from tariff carve-outs, only 20% of electrical power grid buildout equipment receives similar exemptions, doubling the cost per megawatt for combined-cycle gas turbines and delaying critical infrastructure for data centers. Hyperscaler Capex Risks and Custom Chip Competition (39:20) Cembalest draws explicit historical parallels between today's semiconductor indices and the 1999 telecom bubble, where infrastructure equipment makers continued surging even after end-user buyers began rolling over. Major tech hyperscalers have recently shifted from financing immense capital expenditures out of free cash flow to issuing debt, all while free cash flow margins compress. Cembalest warns that within the next 18 months, these hyperscalers must demonstrate that corporate customers are generating sufficient economic returns to justify roughly $1.5 trillion in capital outlays. At the same time, hardware margins face internal pressures. Citing J.P. Morgan Equity Research, the discussion notes that custom in-house accelerators (such as Google TPUs and AWS Trainium) achieve 30% to 40% total-cost-of-ownership savings over commercial GPUs, eroding Nvidia’s pricing power and compressing its valuation multiples. Outside the tech core, late-cycle speculative signals have emerged, including record South Korean retail margin debt piling into semiconductor heavyweights and approximately $500 billion in daily-rebalancing leveraged ETFs amplifying market volatility. AI Disruption, Enterprise Economics, and Patchmageddon (67:10) The economics of enterprise artificial intelligence are changing rapidly. Bridgewater Associates research demonstrated that off-the-shelf, open-weight models fine-tuned on internal proprietary data can outperform frontier commercial models at roughly 5% of the cost. Meanwhile, data from OpenRouter indicates that nearly 50% of U.S. enterprise token traffic has shifted toward low-cost Chinese open-source models like GLM 5.2. Cembalest ranks business model uncertainty highest for consumer-facing providers like OpenAI relative to enterprise-focused Anthropic or infrastructure providers like SpaceX. Beyond market economics, advanced AI models pose severe operational threats. Under an initiative titled 'Project Glasswing,' J.P. Morgan and other major institutions analyzed frontier models and found that AI discovers complex, multi-step code vulnerabilities far faster than linear human developer capacity can patch them. Cembalest warns this widening gap—dubbed 'Patchmageddon'—leaves corporate networks and critical industrial programmable logic controllers highly vulnerable to automated exploitation. While leading indicators like manufacturing new orders relative to inventories suggest the underlying U.S. macroeconomic expansion remains healthy, Cembalest cautions that the convergence of unchecked entitlement spending by 2031, unproven capex returns in AI, and widening cybersecurity vulnerabilities will test long-term market stability.

摩根大通资产管理公司市场与投资策略主席迈克尔·森巴莱斯特（Michael Cembalest）指出，尽管美国经济在创新活力与美元主导权支撑下长期具备韧性，但联邦财政将在2031至2032年逼近利息与法定支出吞噬全部税收的“交叉点”；与此同时，科技巨头依靠举债维系AI资本开支、定制芯片与开源模型的低成本冲击，以及网络安全漏洞的指数级爆发，正在形成不容忽视的结构性风险。 美元储备地位与对美投资的长期实证支撑 (16:32) 森巴莱斯特强调，自2008至2009年全球金融危机以来，摩根大通的财富管理策略始终超配美国资产与美元，这一长期多头策略创造了巨大的跨周期回报。尽管市场频繁出现估值过高的担忧，但考虑到科技等高盈利板块的高增长预期，美股的市盈率相对盈利增长比率（PEG）与全球基准基本一致。在货币层面，美国GDP虽仅占全球约20%，但美元在跨境贷款、外汇储备和SWIFT支付清算中的占比几乎是其经济体量的两倍。关于央行减持美元购入黄金的论调，他澄清黄金在外汇储备中市值占比上升的95%由金价从1500美元上涨至4000美元以上的市值变动驱动，央行实际持有的金衡盎司实物规模持平甚至略降。 针对人民币是否能挑战美元，森巴莱斯特基于货币机制指出，中国M2与GDP比率大幅扩张，加之严格的外汇与资本管制，使其根本无法承担全球储备货币功能。2025年中国估算有8000亿至1万亿美元资本外流，若贸然放开资本账户恐引发不可控的资本外逃。此外，针对中国抛售美债的恐慌，智库外交关系协会学者布拉德·塞策（Brad Setser）的研究显示，中国通过比利时欧洲清算所（Euroclear）离岸托管账户持续购债，实际美债敞口保持稳定；出于消化巨额贸易顺差与获取美元流动性的自身利益，中国无法脱离兼具深度与流动性的美债市场。 2031年财政交叉点与联邦债务倒计时 (32:44) 森巴莱斯特提出了明确的数学推演：到2031或2032年，美国联邦政府将迎来危险的“交叉点”（Crossover Point），届时100%的联邦财政税收收入都将被国债利息支出和法定强制性福利支出消耗殆尽，国防、基础设施建设与政府日常治理将面临无资金可用的境地。他警告称，评级机构不会坐等悬崖降临，极有可能在交叉点到来的两至三年前（即2028至2029年左右）向国会发出正式降级预警或调降主权评级，以迫使其启动类似辛普森-鲍尔斯（Simpson-Bowles）蓝图式的福利结构改革。 尽管二战后艾森豪威尔政府曾在不增税、不削减开支的前提下，依靠高增长将债务占GDP比重从约100%压降至60%，但当前美国面临人口老龄化与贸易摩擦等结构性阻力，难以复刻当年的高增长路径。目前的促增长政策受到贸易关税与能源基建瓶颈的严重制约。例如在AI电力供应生态中，约80%的半导体进口享有免税优惠，但电网建设设备仅约20%获得豁免；联合循环燃气轮机每兆瓦单价在三年内翻倍，部分数据中心甚至被迫串联飞机引擎和船用发动机发电，供应链通胀严重削弱了生产端的扩张效率。 超大规模企业债务转向与半导体周期隐忧 (39:20) 森巴莱斯特对半导体板块权重占标普500指数约18%的现象表达了高度警惕，并指出了其与1999年科技泡沫初期的惊人相似性。在1995至1999年期间，思科等电信硬件供应商伴随互联网服务商（ISP）快速上涨；但到了1999年初，下游买家股价率先拐头下行，而上游硬件设备商却仍在持续扩建产能。当前市场正在重演这一分化：四大超大规模云厂商（Hyperscalers）与甲骨文的股价增长势头开始放缓，半导体指数却持续飙升。如果科技巨头无法证明其1.5万亿美元的庞大资本开支能够产生合理投资回报，位于供应链末端的硬件设备商将面临订单骤降的冲击。 更关键的转折在于科技巨头的资金来源：此前多年巨头均以内部自由现金流支持研发与基建，但去年年底起已转向发行债券融资，导致自由现金流利润率承压。与此同时，英伟达GPU的护城河正面临内部自研ASIC芯片（如谷歌TPU、亚马逊Trainium、微软Maia等）的蚕食。摩根大通股票研究团队测算显示，定制自研芯片相比英伟达方案可降低30%至40%的总体拥有成本（TCO），这也是英伟达远期市盈率回落至约20倍的重要原因。此外，韩国KOSPI指数散户融资买入半导体股票的狂热，往往是科技周期进入极晚期的经典预警指标。 低成本AI竞争冲击与“补丁末日”安全威胁 (67:10) 在模型技术与商业化层面，AI市场正遭遇开源生态与低成本模型的冲击。桥水联合基金（Bridgewater）的研究显示，利用企业内部专有经济与金融数据微调现成的免费开源权重模型，其表现超越了一线前沿大模型，而成本仅为后者的约5%。OpenRouter数据显示，近50%源自美国企业的Token流量正流向智谱GLM 5.2等中国低成本模型。随着各行业CFO对激增的Token账单设置使用上限，以及企业CTO借助AI工具要求传统软件供应商降低订阅费用，过度依赖消费者订阅的商业模式面临考验。森巴莱斯特认为，在未上市独角兽中，OpenAI的商业不确定性显著高于深耕企业级市场的Anthropic与业务壁垒稳固的SpaceX。 除了商业回报，前沿AI模型对数字基建构成了严重的非对称安全威胁。森巴莱斯特透露其即将发布的报告题为《补丁末日》（Patchmageddon），源自摩根大通与数十家大型机构参与的“玻璃翼项目”（Project Glasswing）。测试表明，神经网络能够推演出超越人类程序员思维维度的极端多步攻击链路，使软件与可编程逻辑控制器（PLC）漏洞呈指数级爆发，而人类编写与部署安全补丁的速度仍呈线性增长。这种防御与攻击的不对称性，加之黑箱算法难以向监管机构合理解释决策逻辑，将使全球关键基础设施在未来一段时期面临极高的被渗透风险。 森巴莱斯特认为，尽管制造业新订单对库存比率等先导指标显示美国当前实体生产基础依旧稳健，但在未来18个月内，科技巨头必须向市场兑现数万亿AI投资的商业变现能力；而放眼至2028至2032年，联邦债务跨过财政平衡红线不可避免。技术回报验证与主权债务重构的双重考验，构成了美国未来数年最为核心的宏观挑战。

## Background 本期节目由知名财经播客频道 The Compound 于 2026 年 7 月 13 日在 YouTube 平台发布。节目正值《The Compound and Friends》迎来第 250 期的重要里程碑，主持人 Michael Batnick 与 Downtown Josh Brown 邀请到摩根大通资产管理公司市场与投资策略主席 Michael Cembalest 共同坐镇。根据官方介绍，本期对话深度探讨了美元地位与中国对美霸权的挑战、外国投资者对美债的真实需求、联邦债务的清算时刻，以及剧烈变动的人工智能投资图景、英伟达面临的竞争与半导体的周期性等核心议题。 ## TL;DR Michael Cembalest 基于翔实的宏观数据破除了“美元即将崩溃”与“中国疯狂抛售美债”的悲观叙事，指出美元作为储备货币的统治力依然稳固，但警告美国财政将在 2031–2032 年迎来利息与福利支出吞噬全部税收的系统性“交叉点”。针对当前狂热的科技板块，他将生成式 AI 称为过去 40 年来技术层面最复杂的市场催化剂，直言半导体与云厂商资本开支已显露类似 1999 年思科时期的脱节迹象，同时自研 ASIC 与开源微调模型的崛起正对传统 SaaS 估值与英伟达护城河带来实质性挤压。此外，嘉宾还深入剖析了 AI 带来的网络安全“补丁末日”危机，并从商业模式底层梳理了 OpenAI、Anthropic 与 SpaceX 的业务确定性差异。 ## Atmosphere & style 整场对话在专业硬核的宏观推演与纽约老友式的轻松幽默之间自如切换。面对 64 岁资深策略师严肃沉稳、数据驱动的“冷水式”拷问，两位主持人频繁以机智吐槽与敏锐比喻化解枯燥的技术与财报细节，从“对着 Claude 宣泄情绪”到“外星生命般的黑盒算法”，现场交锋兼具思想张力与笑点。尾声处冷烟火的突袭与全场高呼“250 期快乐”的欢脱场面，更给这场涉及万亿资本与债务炸弹的严肃复盘平添了一抹温暖而鲜活的人情味。 ## Bulletin - [04:15](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=255s) 桥水基金实测显示，经内部专有数据微调的开源权重模型在生成投研简报的质量上反超闭源商业模型，运行成本仅为其约 5%；目前美国开发者近 50% 的 API Token 请求正因极致性价比流向中国大模型。 - [32:44](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1964s) 摩根大通 Michael Cembalest 警示，美国联邦债务将在 2031 至 2032 年迎来交叉点，届时 100% 联邦税收将被国债利息与强制性福利吞没，评级机构预计将在未来 2 至 3 年内向国会下达紧缩通牒。 - [39:22](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2362s) 生成式 AI 正重演 1999 年思科时刻，半导体与基建股持续冲顶但下游超大规模云厂商股价显露疲态，巨头从现金流自给转向发债甚至股权融资，面临在 18 个月内证明 1.5 万亿美元资本开支回报的严峻考验。 - [22:32](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1352s) 全球央行黄金外汇储备占比上升 95% 源于金价飙升带来的价格重估幻象，其实际物理持仓量并未扩张；同时中国官方持债量下滑被比利时 Euroclear 等离岸托管账户的持续买入对冲，并未实质性清空美债。 ## Key takeaways - [01:08](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=68s) AI 在低容错率领域落地门槛依然极高，严肃医疗判读等核心专业环节短期内无法完全实现脱人化，但其低成本图文生成与全天候陪伴属性正在重塑日常交互体验。 - [06:08](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=368s) 开源权重模型正在动摇高定价商业闭源大模型的护城河，中国模型的免费开源策略客观上对依赖高估值的美国 AI 巨头构成了资本维度的降维打击。 - [09:46](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=586s) 在中美 AI 生态割裂与出口管制下，苹果等跨国巨头被迫在不同市场采取双轨制（美国合作 Google Gemini，中国合作阿里大模型），而中国正加快推进光刻机等半导体全产业链自主替代。 - [16:09](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=969s) 自 2008 年金融危机以来，长期超配美股与美元资产的策略几乎不可替代；美股科技巨头表面市盈率高企，但高盈利增速（PEG 视角）为其估值提供了合理性支撑。 - [19:42](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1182s) 尽管美国仅占全球 GDP 的 20%，但美元在 SWIFT 支付与全球外汇储备中仍牢牢占据约 40% 的份额，五年内日元、英镑及人民币均未出现实质性蚕食，做空美元缺乏现实依据。 - [25:21](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1521s) 中国存在高达 8,000 亿至 1 万亿美元的潜在年资本外流压力，加之 M2 规模相对于外储资产超常扩张，严格的资本账户管制决定了人民币短期内无法承担全球终极自由兑换储备货币职能。 - [28:42](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1722s) 美国海外美债占比下降的核心原因是本土发债节奏过快而非外国资金集体离场，中国对美债的实质持仓在计入离岸托管后总体保持平稳。 - [32:44](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1964s) 艾森豪威尔时期的战后超额自然增长不可复制，美国无法单靠经济增速自然化解债务，未来必须通过削减法定福利与增税等紧缩政策进行财政重组。 - [35:36](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2136s) AI 基建正遭遇严重的电力与重型设备硬约束，关税豁免不对称与燃气轮机交付周期长达三年，迫使数据中心转向船舶柴油机甚至退役飞机引擎串联发电自救。 - [38:07](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2287s) 标普 500 指数本质是生产型指数（60%-70% 利润来自商业与科技生产），而美国实体经济则是消费型指数（70% 由消费驱动），两者走势在历史上频繁脱节。 - [47:56](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2876s) 企业端 AI 使用量爆发触发 Jevons 悖论，导致总 Token 支出失控，CFO 介入设立硬性用量上限；同时企业利用 AI 自建作为议价筹码，打破了传统 SaaS 软件每年 3% 的无脑涨价神话。 - [52:49](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3169s) 云厂商自研 ASIC 芯片正在大幅摊薄英伟达的垄断溢价，可为云厂商降低 30% 至 40% 的 TCO；同时双倍杠杆 ETF 的每日被动再平衡机制加剧了美股尾盘的剧烈异常波动。 - [57:26](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3446s) 韩国综合股价指数（KOSPI）散户保证金贷款创下新高，历史上该散户群体的疯狂加杠杆往往是科技板块见顶的典型逆向预警指标。 - [58:58](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3538s) SpaceX 采用分批错峰的阶梯式锁定期成功平抑了解禁抛售波动；主流指数对其纳入出现策略分化，标普坚持持续盈利标准，而纳斯达克先行吸纳。 - [01:01:45](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3705s) 在跟踪的宏观指标中，制造业 PMI“新订单与库存之比”是预测标普 500 企业盈利与股市牛熊走势最精准的前瞻先导指标，目前该比值依然稳健。 - [01:04:54](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3894s) 从底层商业模式的不确定性来看，OpenAI 因高度依赖 C 端订阅且缺乏广告护城河位列最高（面临类似早期 AOL 的流失隐忧），Anthropic 居中，而业务壁垒极高的 SpaceX 确定性最强。 - [01:07:10](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=4030s) 摩根大通深度参与的 Project Glasswing 揭示了“Patchmageddon”网络安全危机：AI 正在以反直觉的人类盲区路径指数级挖掘系统隐秘漏洞，导致企业安全补丁研发陷入严重滞后。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=0s) 闲聊纽约天气与神秘的“鹰头章鱼” 节目在轻松的闲聊中开场。主持人迈克尔·巴特尼克（Michael Batnick）与约书亚·布朗（Downtown Josh Brown）谈起纽约诡异的气候变迁——迈克尔的生日在五月，过去他总能在五月份穿着短裤去钓鱼，如今纽约的五月却格外阴冷；反而是九月、十月的秋季远比春季舒服得多。 随后话题转到摩根大通（JPMorgan）策略师迈克尔·森巴莱斯特（Michael Cembalest）近期报告中使用的奇特封面图：一只半鹰半章鱼的怪兽。迈克尔解释其拉丁语合成词为“Aquilaceph”（结合了拉丁语中代表鹰的 *Aquila* 与代表章鱼/头足类的词根 *Cephalopod*）。布朗询问这幅引人注目的诡异插画是摩根大通内部插画师手绘的还是 AI 生成的。森巴莱斯特透露，摩根大通过去会外包给外部艺术家，但现在直接用 GPT 等大语言模型生成插画效率极高，即便是与一年前相比，模型能够持续叠加细节指令并按要求出图的能力也已经不可同日而语。 ### [01:08](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=68s) AI 的现实边界：从医疗诊断到情绪宣泄工具 当被问及“很多人抱怨 AI 的短板，是否忽略了其演进速度”时，森巴莱斯特给出了务实的回应：AI 在特定领域的进化固然迅猛，但“当犯错的成本极高时，采用的门槛就必须定得非常高”。几年前业内曾断言放射科医生将被 AI 完全替代，但现实中若面临乳腺癌筛查等生死关头，患者依然需要“医生 + AI”，绝不可能完全交付给 AI 独立判断。AI 目前非常适合处理低错误成本的任务（比如画一只鹰头章鱼），但在严肃的医疗判读领域尚无法真正挑起大梁。 现场主持人随即分享了日常使用中的趣味反差： - 巴特尼克笑称自己反而喜欢 AI 不完美，“我最享受对着它发火，我平时不敢对别人说的那些气话全丢给 Claude，它成了最好的减压出气筒，我都担心哪天会被 Claude 封号”。 - 森巴莱斯特则对越来越多人把大模型当作心理医生、人生顾问甚至婚姻咨询师感到意外与费解。 - 布朗提到自己的妻子整天和 ChatGPT 聊饮食营养。过去她每周专门花钱去见营养师汇报饮食，现在她更喜欢随时随地把在意大利餐厅吃了什么拍照或打字输入给 AI，这种“随时在线（Always-on）”且百依百顺的陪伴式体验彻底取代了定期预约的专业服务。 - 森巴莱斯特幽默地补充：即便在曼哈顿找昂贵的专科名医看病，电话里向医生描述症状时，背景音里也能清晰听到医生正对着电脑噼里啪啦敲键盘，然后照着 Google AI Overview 的搜索结果念出来。 专业服务机构未来必须将自身的内部专有知识库转化为随时待命的 AI 交互端，这不是简单的裁员替代，而是专业能力的延伸。 ### [04:15](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=255s) 开源开源权重模型崛起与对冲基金的实测冲击 森巴莱斯特抛出了一个对资本市场极具颠覆性的最新趋势：**采用企业自身专有数据进行微调的免费开源权重（Open-weight）模型，在特定业务场景下的表现正在超越收费高昂的前沿顶尖商业闭源实验室（Frontier Labs）。** 全球顶级对冲基金桥水（Bridgewater Associates）近期发布了一项实测研究：他们将大量隔夜宏观经济数据、新闻报道和研报喂给模型，训练其自动筛选、加权并提炼出最具投资价值的核心信息，为基金经理打包生成清晨阅读简报。结果表明，使用一款完全免费的开箱即用开源模型，在其自有数据上微调后，产出质量竟优于昂贵的 OpenAI 前沿商业模型，而运行成本仅有后者的约 5% 左右。 ```mermaid flowchart TD accTitle: Bridgewater Open-weight Model Workflow accDescr: Diagram illustrating Bridgewater's internal AI pipeline where raw data is fed into an open-weight model to produce morning briefings at 5% cost. subgraph Inputs["数据输入端"] D1["隔夜经济数据发布"] D2["突发财经新闻资讯"] D3["外部卖方研究报告"] end subgraph Processing["模型处理与权重评估"] M["免费开源权重模型<br/>(经桥水自有数据微调)"] end subgraph Outputs["输出与收益"] O1["晨会专属投资研报与资讯包"] O2["综合成本仅为 OpenAI 前沿模型的约 5%"] O3["关键投资信息加权精准度反超闭源商业模型"] end D1 --> M D2 --> M D3 --> M M --> O1 M --> O2 M --> O3 ``` 这一性价比断层直接反映在开发者调用层面。据 Open Router 近期 48 小时内的统计数据，由美国公司产生的 API Token 请求中，有接近 50% 正在流向中国开发的大模型，而非美国本土的前沿商业模型。核心驱动力正是极高的“单 Token 成本性能比（Cost-per-performance）”，这在半年前甚至一年前都是难以想象的转变。 ### [06:08](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=368s) 中国开源大模型的低价策略与地缘商业博弈 针对“为什么中国模型更高效、调用量暴增”的疑问，嘉宾们展开了深度剖析。中国最新发布的开源大模型（如智谱 GLM-4 / 5 系列等）在 GPQA Diamond、Chatbot Arena ELO 评分、数学推理和代码能力等评测基准上，距离 GPT 前沿顶尖版本的差距已缩小到普通用户几乎无法察觉的地步，但其成本和开放度优势极其显著。 现场探讨了这背后的核心驱动与战略意图： - **资本与估值层面的精准打击**：当前美股市场中最值钱的企业要么是 OpenAI、Anthropic 这样的前沿模型实验室，要么是向其贩售算力硬件的巨头（如英伟达）。如果中国投入数十亿美元研发出几乎同等水平的模型并免费向全球开源，就是在“给这个巨大的估值气球扎针”，通过极限压低大模型的商业化溢价，直接削弱美国科技巨头赖以维持超高估值的护城河。 - **算力生态与本端部署**：这些开源权重允许全球企业直接在本地服务器或搭载英伟达芯片的本地设备上离线运行，数据完全不需要传回中国服务器，消除了企业的数据合规顾虑，同时培育了对自身架构的生态依赖。 - **商业模式的历史重演与监管隐忧**：布朗对比了早期互联网与社交媒体的免费逻辑——“如果你不付钱，你就不是客户，而是产品本身”。森巴莱斯特则预判，随着中国高性能开源模型在全球开发者中的广泛蔓延，美国商务部未来很可能会像欧洲应对席卷汽车市场的中国低价电动车（EV）一样，开始审查甚至限制中国 AI 模型的跨境扩散风险。 ### [09:46](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=586s) 中美 AI 生态的分野与自主突围 对于中美在人工智能领域的竞争，嘉宾坦言技术复杂性已经远超以往的任何科技革命：“在历史上所有的技术革命中，这一场是最难理解各个细枝末节的。”相较于早期的光纤网络铺设、互联网普及以及 B2B 软件潮，当前 AI 的底层构架要复杂得多。 业内普遍承认，中国在大模型上的部分进展确实借鉴了用美国大模型生成数据进行训练的方法，但这绝非全貌。面对美国的严格出口管制，中国正在被迫且坚定地打造属于自己的闭环生态系统。例如，中国正在研发能够替代荷兰 ASML（阿斯麦）的光刻机，虽然目前的技术水平与顶级设备仍有差距，但正在快速迭代追赶。 这种生态割裂在跨国科技巨头的商业落地上尤为明显： - **苹果的“双轨制”策略**：苹果在美国的 AI 合作伙伴选定为 Google 的 Gemini 模型；但由于 Google 服务无法在中国运作，而中国又是苹果极其庞大的核心市场，苹果只得转而与阿里巴巴达成合作，采用阿里的大模型。这使得苹果在不同市场形成了两套完全割裂的 AI 运行生态。 --- ### [11:30](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=690s) 第 250 期里程碑与赞助赞助商致辞 节目迎来了《The Compound and Friends》第 250 期的重要里程碑，主持人 Michael Batnick 与 Downtown Josh Brown 共同庆祝这一特殊时刻，并向幕后团队致谢。 随后进行了本期赞助商 Federated Hermes 的介绍：面对日益复杂的市场环境，主动型 ETF 正在重塑投资组合的构建方式。Federated Hermes 推出了涵盖股票与债券市场的 12 只主动型 ETF 系列，强调在严苛的市场压力测试与推演下筛选标的，兼具核心配置功能与战术调仓的灵活性。 --- ### [13:06](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=786s) 传奇策略师 Michael Cembalest 与“理性爱国主义” 节目正式迎来重磅常驻嘉宾——摩根大通资产管理公司（J.P. Morgan Asset Management）市场与投资策略主席 Michael Cembalest。他冒雨从第 47 街步行赶到录音室。Cembalest 自 1987 年加入摩根大通，曾担任该机构全球私人银行长达八年的首席投资官，其长期撰写的旗舰市场洞察刊物《Eye on the Market》广受全球机构投资者关注。 主持人 Josh Brown 饶有兴致地探问 Cembalest 是否算得上一个“爱国者”。Cembalest 明确表示自己确实是一个爱国者，但强调自己的爱国情怀建立在“数据分析”之上： - **人口流动印证国家吸引力**：世界银行（World Bank）的跟踪数据显示，美国依然是全球移民迁入与迁出比率最高的国家，全球人才在“用脚投票”。 - **隐忧依然存在**：即便看好美国基本盘，Cembalest 也指出了长期风险。除了不可持续的联邦债务水平外，他对法治环境（rule of law）的受侵蚀以及联邦政府对医学基础科研资助的现状同样抱有深切顾虑。 --- ### [16:09](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=969s) 为何做空美国总是失败？“半五百年触手”与美股溢价 Cembalest 将其最新一期的《Eye on the Market》特别报告命名为《Semiquentacles》（将美国建国 250 周年 Semi-quincentennial 与触手 Tentacles 融合的自造词），生动形象地表明：尽管外界数十年来唱衰不断，美国的资本与制度“触手”依旧紧密维系着全球市场的主导权。 ```mermaid flowchart TD accTitle: 长期超配美国资产的投资逻辑架构 accDescr: 展示自2008年以来做多美国资产的构成要素、对冲日本泡沫的历史对比以及面对高估值时的PEG盈利增长支撑逻辑。 subgraph CoreStrategy["超配美国核心资产组合 (2008/2009 至今)"] direction TB A["美元资产 & 美债信用"] B["美股核心资产"] C["美国商业地产与基础设施"] end subgraph Valuation["估值争议与基本面反驳"] direction TB D["表面现象: 市盈率 (P/E) 处于高位"] --> E["深层支撑: 高盈利增长 (Earnings Growth)"] E --> F["综合考量: PEG 指标与其他市场相当"] end subgraph HistoricParallels["历史收益对比"] G["长期做多美国资产的惊人复合回报"] --- H["历史上唯一能与之匹敌的策略: 1989年低配暴跌前的日本股市 (曾占全球65%)"] end CoreStrategy --> HistoricParallels CoreStrategy --> Valuation ``` 自 2008 年至 2009 年 3 月金融危机底部以来，摩根大通建议客户全面超配美元、美国信用债、美股、美国房地产以及基础设施，这一策略带来了极为丰厚的回报。在 Cembalest 看来，过去 15 到 18 年里，唯一在超额收益上能与“做多美国”相提并论的宏观交易，只有在 1989 年日本股市占据全球市值 65% 的巅峰期时选择“坚决低配日本”。 针对市场动辄高呼“美股见顶”、“估值过高”的论调，Cembalest 与主持人展开了辩护： - **估值与增长匹配（PEG 视角）**：粗看表面市盈率（P/E）确实昂贵，但估值最高的板块恰恰兑现了最强劲的盈利增速。将市盈率与盈利增长率结合考量（PEG Ratio），美股科技板块相较全球其他市场并未出现离谱的脱节。 - **经典估值悖论**：正如华尔街流传的一句名言：“我不需要花钱请分析师告诉我 8 倍市盈率的股票便宜；我需要的是有人告诉我，为什么 40 倍市盈率的股票反而是便宜的，而 8 倍市盈率的却可能极其昂贵。” - **缺乏替代者的美元霸权**：许多看空者抨击美国的巨额财政赤字、债务与 GDP 占比飙升以及不可持续的货币政策，却从未能指出真正的替代者。Cembalest 指出，虽然终有一天美国需要为债务“买单”，但鼓吹人民币或所谓“中国方案”在现阶段取代美元储备货币地位，从现实逻辑来看是完全荒谬的。 ### [19:42](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1182s) 美元霸权并未动摇：看空者的逻辑盲区与资产配置定力 摩根大通的 Michael Cembalest 阐明了其团队管理资金的核心纪律：机构旗下管理着代表机构与个人投资者的 4 万亿美元巨额资产，首要责任是避免“过早逃跑”，绝不能在牛市中途因恐慌退场而把巨额收益拱手让人。市场上常年充斥着“美元崩盘在即”的末日论调，但 Cembalest 强调，严谨的投资者必须用客观数据说话：美元在全球贸易结算份额、外汇储备占比、SWIFT 国际支付结算份额以及跨国银行贷款等核心维度上，究竟有没有出现任何崩盘迹象？ 数据显示，尽管美国本土经济仅占全球 GDP 的约 20%，但美元在上述各项全球指标中的份额均是该比例的两倍左右，展现出绝对的垄断统治力。Cembalest 指出，很多评论家喜欢以鲁比尼式（Nouriel Roubini）的极端悲观调门猛烈唱衰美元，却缺乏基本的实证纪律，从不肯正视美元作为世界储备货币的核心支柱根本没有发生任何动摇。图表显示，在过去五年中，日元、英镑、人民币等主要货币的全球份额均处于横向震荡，没有任何单一货币能蚕食美元的地位；挪威克朗或新加坡元等偶尔迎来极小规模配置的货币，体量更属于“微型货币”，完全不具备充当储备货币的容量。因此，机构在资产配置上绝不会对投资组合的底层货币覆盖进行激进的结构性调整。 主持人们进一步剖析了美元看空者的典型误区。看空者常拿“美元购买力在过去半个世纪蒸发了 96%”作为论据，但这本质上是通胀环境下的同义反复——以其他法币计价，物价同样在上涨。更荒谬的是，这批高喊美元购买力缩水的人，往往又同时劝阻大众参与股票市场。事实上，现金绝不能被锁在床垫底下任由通胀蚕食；抵御法币购买力贬值的有效工具正是股市与优质房地产，美股长期的复合回报已大幅跑赢通胀，完美维护了资金的实际购买力。 - **管理规模与定力**：Cembalest 团队负责监管 4 万亿美元资产，必须抵制盲目退场的恐慌情绪。 - **超配现实**：美国仅贡献全球约 20% 的 GDP，但美元在 SWIFT 支付、外汇储备与跨境贷款等领域均占据约两倍于此的压倒性份额。 - **竞争币种受限**：日元、英镑、人民币五年内均无显著份额扩张，小型主权货币容量微弱，无法承担储备职能。 - **购买力悖论**：单纯持有现金确实面临购买力损耗，但“唱空美元”与“看空美股”自相矛盾，股票与地产是抵抗购买力流失的终极资产。 --- ### [22:32](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1352s) 主权基金对数字资产的考量与央行购金背后的“计价幻象” 面对近年来美国宏观基本面的争议，部分避险资金流向了黄金与加密资产。Cembalest 指出，判断资产属性是否发生质变的关键裁判标准，是新加坡政府投资公司（GIC）、淡马锡（Temasek）等重量级主权财富基金，是否会在外汇储备中将数字资产从短期的“战术性交易”提升为长期的“战略性配置”；而迄今为止，这种根本性的范式转移并未发生。与此同时，美国政治与货币政策的博弈也在牵动市场神经。特朗普曾公开呼吁在年底前将联邦基金利率压低至 1% 以下，这种行政力量对央行独立性的干预，令人联想起当年尼克松总统与美联储主席阿瑟·伯恩斯（Arthur Burns）之间的对抗剧本。然而，被提名为美联储官员的凯文·沃什（Kevin Warsh）公开发言表明其坚持稳健货币（hard money）的强硬立场，绝非轻易妥协之人。预测市场 Kalshi 上显示年内加息概率依然维持在 53% 左右，与他上任前基本持平，而沃什展现出的鹰派姿态也直接给此前躁动的黄金与加密货币行情降了温。 ```mermaid flowchart TD accTitle: Central Bank Gold Reserve Value Surge Diagram accDescr: A comparison showing gold price appreciation versus actual physical troy ounces held by central banks. subgraph Metric1 ["名义储备价值暴增 (Valuation)"] direction TB P1["金价从 1500 美元飙升至逾 4000 美元/盎司"] --> P2["央行外汇储备中黄金金额占比陡峭上扬"] end subgraph Metric2 ["实际物理持仓 (Physical Weight)"] direction TB W1["央行实际持有的黄金物理重量 (盎司/吨位)"] --> W2["物理持仓持平甚至略微回落"] end subgraph Metaphor ["马克·安德森房产隐喻"] direction TB M1["初始：10万美元现金 + 10万美元房产 (各占50%)"] M2["知名投资人出价100万美元收购该房产"] M3["房产名义配置比例升至约90% (非主动增持，纯因资产重估)"] M1 --> M2 --> M3 end ``` 谈及备受热议的“全球央行爆买黄金”叙事，Cembalest 用一个通俗的比喻戳破了统计假象：假设某人有 10 万美元现金和一栋价值 10 万美元的房子；某天风险投资家马克·安德森（Marc Andreessen）突然出价 100 万美元要买这栋房产，此时该业主的资产账面结构瞬间变成了 90% 是房产。但这仅仅是价格重估（price change）导致的账面比例上升，并不意味着他主动增购了更多不动产。市场上各类媒体大肆渲染“央行外汇储备中黄金占比飙升”，其实 95% 的原因在于金价从 1,500 美元一路暴涨至 4,000 美元以上；如果从央行实际持有的黄金物理盎司总量来看，这一数字不仅没有激增，甚至还出现了下滑。所谓“央行加速扫货囤金”的观点，很大程度上是黄金死多头（gold bugs）对宏观数据的片面误读与夸大宣传。 - **主权基金配置底线**：大型主权基金尚未把数字货币纳入正式的外汇战略储备。 - **政治施压与货币独立性**：尽管面临将利率骤降至 1% 的政治施压呼声，沃什的“硬货币”鹰派表态让美联储年内加息的预期保持在 53% 左右，直接遏制了贵金属与加密市场的狂热情绪。 - **黄金占比增加的真相**：央行外汇中黄金名义比重走高主要受金价上涨的被动推升驱动，实际储备的物理黄金盎司数并未扩容。 --- ### [25:21](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1521s) 货币堰塞湖与汇管壁垒：为何人民币尚难取代美元？ 讨论转向中国经济对美元体系的潜在竞争。Cembalest 承认中国在诸多科技创新、专利产出和工业实验领域取得了惊人成就，其赶超步伐毋庸置疑；但在金融制度设计上，中国存在着严密的人身与跨境资金流动管控。据估计，即便在严密的外汇管制下，中国仅在 2025 年记录在案的资本外流规模仍高达 8,000 亿至 1 万亿美元。如果中国全面开放资本账户，势必引发天量资金外逃，从而直接摧垮人民币汇率并重创国内的股票与房地产市场。 为了直观呈现这种结构性矛盾，Cembalest 引用了童年经典节目《芝麻街》（Sesame Street）中“哪一个与其他不同”（one of these things is not like the other）的比喻，将全球三种经济体在“货币供应量、央行资产与外汇储备”的关系上进行横向对比： 1. **盯住汇率制经济体**（如中国香港特区、新加坡以及部分海湾产油国）：严格匹配货币供应与外汇储备，通过精准调节维持汇率挂钩稳定； 2. **浮动汇率与自由流动经济体**（如墨西哥比索体系）：货币发行与资产负债表保持市场化匹配； 3. **中国模式**：广义货币（M2）相对 GDP 的比例极其惊人，其 M2 供应量的超常扩张远远甩开了央行资产与外汇储备的积累斜率。 这种海量被“圈禁”在境内的流动性形成了一座巨大的货币堰塞湖，如同一柄达摩克利斯之剑悬在头顶。一旦外汇闸门松开，庞大的民间财富势必疯狂涌向海外寻求多元化配置（例如流入温哥华豪宅或全球美元资产）。世界储备货币的核心特征之一是具备完全自由兑换的资本流动性，因此在这种资本账户壁垒和潜在外流压力彻底解决前，人民币并不具备充当全球终极储备货币的制度前提。针对 Batnick 提出的“为何如此惊人的货币供应量没有在境内引发恶性通胀”的疑问，讨论指出，这是因为房地产市场的深度调整引发了强烈的资产负债表收缩与通缩压力，加上居民储蓄意愿高企，大量资金沉淀在固定收益产品与银行储蓄中，有效对冲了 M2 激增可能带来的物价通胀效应。 - **资本外流潜在风险**：即便存在严格监管，2025 年中国年资本外流规模仍据估计达 8,000 亿至 1 万亿美元；全面放开资本管制或将引发汇率与资产价格剧烈震荡。 - **M2 与外储脱节**：中国的货币供应扩张速度显著超越了央行外储资产的积累速度，资本账户的非自由兑换构成了其走向全球储备货币的核心制度障碍。 - **通胀抑制机制**：境内房地产价格的下行压力与高额的居民储蓄倾向吸收了过剩的流动性，阻断了货币供应向消费物价通胀的传导链条。 ### [28:42](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1722s) 中国“抛售美债”的迷思与美元资产的真实流动 主持人与嘉宾首先谈到了中国经济的内部结构与高储蓄率问题。中国当前的社会保障机制与百年前的美国类似，在医疗保障等公共安全网方面仍然相当薄弱。随着人均预期寿命延长而制度性安全网尚未健全，居民不得不采取极高的预防性储蓄。这种经济特征随后引申出了市场关注的核心议题：所谓中国正在大规模“抛售美国国债”甚至罢买美债的传闻。过去深夜脱口秀节目经常调侃“中国买下了美国、美国欠了中国巨债”，如今舆论风向却反转为“中国正在疯狂清空美债，他们究竟知道什么我们不知道的内幕”。 Michael Cembalest 指出，投资者猜测全球买家可能对现届美国政府的财政或对外政策产生抵触，这种逻辑看似合理，但客观数据并不支持这一论断。他强调自己不做道德或价值预判，只看真实数据。图表显示，海外投资者仍在以以往基本持平的速度买入美国国债（Treasuries）、短期国库券（T-bills）、企业债及美股。海外持有美债比例看似下降的根本原因并非外国人在大举离场，而是美国发债的速度实在太快。全球范围内真正明确宣布实行“买家罢工”并抛售美债和机构债的只有俄罗斯。至于中国，美国外交关系协会（CFR）的高级研究员布拉德·塞策（Brad Setser）曾深入研究这一问题：中国官方报告的直接持债量虽有下降，但实际上中国通过比利时欧洲清算银行（Euroclear）账户购买了极大量的美债。两项综合计算后，中国的实际美债头寸基本保持平稳。 之所以出现这种情况，核心在于中国别无选择。中国对外贸易高度依赖美元结算，且长期保持巨大的资本账户盈余，海量资金必须有去处。美债市场必须留心这种海外买家结构，因为大家都不希望看到中国突然甩卖数千亿美元资产、迫使美联储或财政部效仿欧洲银行系统那样强压本土商业银行接盘成为“兜底接盘侠”。目前产油国的主权财富基金虽然也在入场，但购买节奏未见井喷，总体上美国依然凭借“美元全球储备货币”享有着举借巨额赤字而无须担心国债无人问津的“过度特权”（Exorbitant Privilege）。 - **预防性储蓄成因**：由于缺乏类似美国 Medicare 和 Medicaid 的完善医疗养老保障网，中国家庭必须保持高储蓄率以应对预期寿命延长带来的长期风险。 - **美债真实买盘路径**：布拉德·塞策的追踪研究证实，中国官方持仓数据的下滑被比利时 Euroclear 等离岸托管账户的持续净买入所对冲。 - **资本账户现实**：庞大的经常项目与资本盈余让中国面临全球范围内缺乏可替代流动性资产池的现实，使得继续配置美债成为维持美元贸易流动性的必然举措。 --- ### [32:44](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=1964s) 2031-2032 债务交叉点：无可避免的债务定时炸弹与增长幻觉 谈到美国联邦债务的终局，Michael Cembalest 提出了一个具体且令人警醒的时间表。他认为真正的系统性“清算时刻”并非突发黑天鹅或无声无息的缓慢侵蚀，而是在 2031 年至 2032 年左右到来的关键“交叉点（Crossover Point）”：届时美国联邦政府的全部税收收入将被 100% 用于支付国债利息和法定福利支出（Entitlements），公共财政将没有哪怕一分钱留给国防、基础设施等任何其他日常政府开支。 ```mermaid flowchart TD accTitle: 联邦财政交叉点危机传导链 accDescr: 展示美国联邦债务增长如何逼近税收枯竭节点并倒逼评级调降与福利改革 subgraph Timeline["时间节点推进"] T1["当前阶段 (2026-2028)"] T2["评级预警期 (约 2028-2029)"] T3["临界交汇点 (2031-2032)"] end subgraph Pressure["财政收支结构"] P1["债务利息激增 + 法定福利支出膨胀"] P2["联邦税收收入 100% 被利息与福利耗尽"] end subgraph Action["政策反应与抉择"] A1["评级机构施压：威胁剥夺主权信用"] A2["重演辛普森-鲍尔斯蓝图：削减福利与增税"] end T1 --> P1 P1 --> T2 T2 --> A1 A1 --> T3 T3 --> P2 P2 --> A2 ``` Michael Cembalest 预计，在这个绝对临界点到来的前两到三年（即大约 3 年后），三大国际评级机构就将以大幅调降美国主权信用评级为威胁，强迫国会进行严肃的福利体系重组。二十年前“辛普森-鲍尔斯（Simpson-Bowles）委员会”就已经为解决债务赤字拟定了成熟政策蓝图，能够真正撬动财政大局的手段屈指可数，无非是增税与削减刚性开支。尽管市场上总有人幻想可以单凭经济高增长“跑赢债务”，就像艾森豪威尔执政时期债务占 GDP 比重从约 100% 暴跌至 60% 那样，但艾森豪威尔接收的是二战击溃德意日法西斯后遗留的特殊战时债务，且当时伴随着极高的名义与实际经济增速。反观当下，无论是保守派承诺但最终滑向本土保护主义的经济路线，还是可能将资源分散于其他社会议题的民主党政策，都不太可能单靠自然增长轻松化解当前的债务死结。 - **债务吞噬财政**：2031–2032 年左右，联邦税收预计将全额被国债利息与强制性福利吞没，赤字率将失去自我修复能力。 - **评级倒计时**：预计 2028–2029 年评级机构就会向国会下达最后通牒，倒逼未来数个立法周期出台实质性财政紧缩案。 - **历史对比的局限**：艾森豪威尔时期的降杠杆依赖战后黄金期的内生超额扩张，无法作为现代和平时期常态化财政赤字自愈的模型。 --- ### [35:36](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2136s) 政策抵消效应与 AI 电力基建的供给侧硬约束 讨论转向当下的经济现实与增长阻力。尽管推行了大力度的监管松绑，美联储此前也做出了政策应对，加上 2017 年减税法案的延续，美国 GDP 增速依然徘徊在 2.8% 左右。Michael Cembalest 解释称，去监管化本身是一个极为漫长的制度释放过程，以 20 世纪 90 年代克林顿政府时期的电信和航空业松绑为例，私营部门通常需要三到五年的调整期才能真正革新商业模式。更关键的是，现行政策内部充满了严重的“自相抵消”：一方面推进去监管，另一方面却在通过高关税推高经营成本并收紧劳动力供给。 这种结构性矛盾在人工智能浪潮的电力配套建设中表现得尤为剧烈。半导体芯片产业链中有约 80% 享受到了关税豁免，但下游电力基础设施建设中仅有 20% 能获得同等待遇，直接导致变压器等核心电网组件价格飙升。Michael Cembalest 形容自己脑海中也有一个跳动的“赤字时钟”，那就是联合循环燃气轮机（Combined Cycle Gas Turbine）的每兆瓦报价——该价格在过去三年翻了一番，且目前仍在持续攀升，交付周期更是长达三年之久。 全球具备大型燃气轮机制造能力的供应商基本仅限于 GE Vernova、三菱（Mitsubishi）与西门子（Siemens），由于产能不能迅速扩容，市场甚至出现了考虑引用《国防生产法》（Defense Production Act）强制其扩大制造的讨论。在极度紧缺的供应环境下，开发商与数据中心运营方为了解决燃眉之急，不得不四处搜罗芬兰瓦锡兰（Wärtsilä）的重型船舶发动机以及退役航空发动机，像“菊花链（daisy chains）”一样串联拼凑起来发电自救。 - **关税豁免不对称**：AI 半导体硬件享受约 80% 的关税除外豁免，而电网与能源基建关键设备仅有 20% 豁免，直接卡死了能源扩张瓶颈。 - **燃气轮机价格与周期暴涨**：联合循环燃气轮机单位每兆瓦成本三年翻倍，订货等待排期长达三年。 - **替代发电极端举措**：由于 GE Vernova、西门子与三菱产能受限，市场被迫转向采购瓦锡兰船舶柴油机及报废飞机引擎进行离网串联发电。 ### [38:07](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2287s) 生产指数与消费指数：标普500与实体经济的脱节 对话开篇探讨了当前美国经济与股市之间的奇特分歧。市场此前大幅推高了 Bloom Energy 等能源相关股票，原因是其制造的固体氧化物燃料电池（solid oxide fuel cells）能够将天然气高效转化为电力，精准契合了数据中心爆发的电力缺口。然而，主持人指出一种普遍的困惑：许多美国上市公司公布的财报利润同比大增 70%、80% 甚至 100%，但在整体宏观经济数据中，却完全看不到同等强劲的势头。 Michael Cembalest 从经济学底层逻辑解释了这一割裂现象。他指出，美国国内生产总值（GDP）大约 70% 是由居民消费驱动的，无论经济周期如何，消费的增速都相对平稳；然而，标普 500 指数中 60% 至 70% 的利润来源于工业、科技和商业生产。简而言之，标普 500 本质上是一个“生产指数”（Production Index），而宏观经济则是一个“消费指数”（Consumption Index）。两者之间的差距完全可能长期持续。历史上也屡见反向脱节：例如在 2001 年互联网泡沫破裂期间，美国实体经济并未出现深重衰退，甚至美国全国经济研究所（NBER）在数年后才勉强确认当时发生过轻微衰退，但标普 500 指数却暴跌了 40%。这种市场与实体经济周期的脱钩并非新鲜事。 --- ### [39:22](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2362s) 半导体周期悖论与“思科时刻”的历史隐忧 主持人和 Michael Cembalest 转向了半导体板块的狂热现象。根据 Joey Politano 制作的图表，美国目前从台湾地区的进口额历史上首次超过了从中国大陆的进口额，这在几年前是完全不可想象的。与此同时，半导体在标普 500 指数中的权重已攀升至约 18%，美光科技（Micron）等内存巨头的市值甚至一度逼近 Meta。主持人提问：这些在历史上具有极端周期性的半导体股票，其周期性是否已经被彻底消除了？ ```mermaid flowchart TD accTitle: 基础设施投资脱钩模型与1999年思科时刻对比 accDescr: 展示1999年与当前AI周期中基础设施提供商与终端买家之间的股票走势背离及风险传导机制。 subgraph TechBubble ["1999年 互联网泡沫前期"] A["下游买家: ISP / 电信公司<br/>(股价率先见顶下跌)"] -->|削减资本开支| B["上游基建: Cisco 等设备商<br/>(股价滞后暴跌)"] end subgraph AICycle ["当前生成式 AI 周期"] C["终端巨头: 四大超大规模云厂商 + Oracle<br/>(面临 $1.5万亿 资本支出回报质疑)"] -.->|若投资回报不及预期| D["上游芯片: GPU / HBM 内存厂商<br/>(目前股价仍在加速冲顶)"] end ``` Michael Cembalest 坦言，大约三个月前他对此并不担心，但近期开始感到焦虑。他回忆了 1995 年至 1999 年的互联网行情：当时网络设备股与电信服务股原本同步上涨，但在 1999 年初出现了一个极度危险的背离信号——思科（Cisco）等基础设施设备股仍在不断飙升，而其下游买家（如互联网服务提供商 ISP、Verizon 等电信公司）的股价却已经开始调头向下。当时包括著名基金经理 Gary Brinson 在内的顶尖分析师就敏锐地警告：下游买家的股价暴跌意味着投资者在惩罚其庞大的资本开支，他们不可能在次年继续购买同等规模的硬件。 如今的历史正在惊人地重演：半导体指数仍在不断加速冲顶，但下游的四大超大规模云计算公司（Hyperscalers）外加甲骨文（Oracle）的股价却已显露疲态。如果这些科技巨头无法向市场证明其高达 1.5 万亿美元的资本支出能够产生合理的回报，上游的基础设施与半导体股票迟早会面临巨大风险。Josh Brown 生动地比喻道：上游组件和光学器件供应商往往是整列火车里“最后拿到备忘录”的车厢，车头的引擎早就撞毁跌下悬崖了，坐在末节车厢的供应商看着眼前堆积如山的订单簿，还误以为形势大好，甚至在彻底崩塌前夕还会宣布大规模扩产。 - 美光科技近期宣布在纽约州北部建设巨型晶圆厂，这与台积电在亚利桑那州的建厂项目类似，很大程度上是地缘政治驱动而非单纯的商业考量。 - Michael Cembalest 赞赏了美国政府强力施压外国企业对美进行直接投资（FDI）的做法，认为这补齐了美国长期以来的政策短板。 - 针对苹果公司游说白宫允许采购中国产内存的问题，Cembalest 指出，在 2 纳米以下尖端半导体及高带宽内存（HBM）领域，顶尖厂商拥有无可比拟的规模经济壁垒，短期内全球内存供应链根本不可能出现突然的供给过剩。 --- ### [44:08](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2648s) 报告免责声明背后的狂热社群与迈克尔·伯里的会计质疑 主持人好奇地提到，Michael Cembalest 在其撰写的 AI 专项投资报告中特意加了一段火药味十足的免责声明，警示读者该报告包含大量技术术语、未经证实的推论，并直言“如果你对这些话题抱有弥赛亚般的狂热信仰，或者无法容忍任何异见，请勿阅读”。Cembalest 笑着解释这是为了提前堵住一些暴躁读者的嘴。 他指出，自己涉足金融行业 30 年，覆盖过包括能源在内的各类激烈博弈板块，但从没见过任何一个领域的信徒像 AI 拥趸那样充斥着狂热、毫无保留且极度片面的情绪。无论是在讨论 Token 价格还是杰文斯悖论（Jevons' paradox），这群主要活跃在 Twitter 等半匿名社交网络上的人总是在对他人尖叫宣泄，极度缺乏面对面深度交流的能力。与此同时，知名空头迈克尔·伯里（Michael Burry）近期提出的质疑也非常值得关注：科技巨头们采购的 AI 硬件究竟被按照多少年进行折旧记账？这些资产的摊销年限是否与其实际的有用使用寿命（useful lives）严重脱节？这些直击财报质量的硬核质询，反映了当前市场情绪在极度亢奋中潜藏的会计盲区。 --- ### [46:12](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2772s) 40 年来技术上最复杂的催化剂与市场的剧烈反转 Michael Cembalest 将当前的生成式 AI 浪潮定性为“过去 40 年来技术层面最复杂的市场催化剂”。他对比了以往的科技周期：在过去 30 年里，无论是企业级软件（B2B Software）还是光纤网络建设，一个拥有初中文化水平的人花点时间就能完全理解其运作原理。然而面对今天的 AI，投资者必须从 2018 年谷歌那篇具有奠基意义的 Transformer 架构论文（《Attention Is All You Need》）开始啃起。 理解现代 AI 产业链需要极高的技术门槛：GPU、TPU 与传统 CPU 的能耗与算力差异，DRAM 与 HBM 的制造物理极限，不同模型的底层工作机制等，都异常深奥。每当有初创公司或巨头发布新的技术突破声明时，专业分析师需要耗费大量精力去拆解验证其真实性。然而，金融市场根本没有耐心等待严密验证，往往凭第一反应草率定价，导致预期发生频繁而剧烈的“180 度大反转”。例如，市场最初曾笃定 OpenAI 会迅速吞噬谷歌的核心搜索与广告业务，使谷歌彻底沦为输家；然而不久后市场又全盘推翻了这一论调，转而认为 Gemini 模型更具优势，且谷歌庞大的全球用户基础能让其比任何独立初创企业都更快地实现商业变现。市场的极度浮躁与技术的极度深奥，构成了本轮 AI 投资狂潮中最显著的矛盾。 ### [47:56](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2876s) 超大规模云厂商的资本开支隐忧与企业端 AI 成本现实 Michael Batnick 与 Josh Brown 谈到人们常因对技术演进路径过度僵化而错失机会。例如过去欧洲曾因反垄断要求苹果手机取消 Google 作为默认搜索引擎的预设，逼迫用户自行选择，但最终绝大多数用户依旧主动选择 Google；现实中几乎没人会说“让我去 Bing 一下”或“去 DuckDuckGo 搜索”。然而，这种超预期的市场演进如今正逼近财务瓶颈。Michael Cembalest 指出，过去几年大型 Hyperscalers（超大规模云厂商）完全依靠自身产生的充沛自由现金流来支撑巨额资本开支（CapEx），但从去年底开始，他们已开始举债融资。尽管从债务与 EBITDA 比率来看目前尚属稳健，但一阶导数趋势已经转向，研报图表显示各大云厂商的自由现金流（FCF）利润率正出现陡峭下行。 在资本开支与研发费用占收入比重持续飙升、现金流快速恶化的背景下，云厂商面临着在 18 个月内拿出商业化真金白银的巨大考验。他们必须证明像摩根大通（JPMorgan）这样的大型机构不仅在加速采用 AI，而且愿意持续支付高昂溢价，而不是转头寻求免费的开源或中国模型。Cembalest 透露，摩根大通内部正针对不断暴涨的 Token 调用成本频繁召开 CFO 会议。Josh Brown 补充分析，虽然单个 Token 的生成成本在持续降低，但整体需求增长呈爆炸式喷涌（Jevons 悖论：效率提升反而刺激了更大的总需求），一旦员工掌握这些工具，只会疯狂探索更多用途，导致许多企业被迫对员工的 AI 用量设立硬性上限。不过大家一致认为，这绝不同于当年的 3D 打印热潮，AI 不可能再被塞回潘多拉魔盒，尤其在 Agentic AI（智能体）迅猛演进并吞噬海量 Token 的当下，这种产业趋势是不可逆的。 ```mermaid flowchart TD accTitle: Hyperscaler Capital Dilemma and Token Cost Squeeze accDescr: Diagram illustrating the shift from organic cash flow to debt financing alongside escalating corporate token usage and CFO pushback. subgraph Hyperscalers["超大规模云厂商财务压力"] A["巨额 CapEx 与 R&D 开支"] --> B["从自由现金流自给转向发债融资"] B --> C["自由现金流 (FCF) 利润率陡降"] C --> D{"18个月验证期: 企业是否买单?"} end subgraph Enterprises["企业端 AI 现实博弈"] E["Agentic AI / 复杂工具落地"] --> F["单 Token 成本下降但总需求暴增"] F --> G["CFO 介入: Token 成本账单失控"] G --> H["实施用量硬封顶 / 评估自研与平替"] end D <--> H ``` ### [51:09](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3069s) 传统 SaaS 的议价危机与私募股权软件神话破灭 AI 的渗透直接冲击了传统软件业延续多年的商业叙事。以往企业面对企业软件（SaaS）合同续约时，几乎都是闭着眼睛接受每年 3% 左右的自动涨价条款（escalator）。如今情况发生 180 度大转弯：每一位企业首席技术官（CTO）都被管理层盘问，为什么不能利用 AI 工具替代部分外部供应商？即使无法彻底替换，CTO 也必须致电供应商，以“自建或转投 AI”为谈判筹码，逼迫对方将订阅费用减半。 这种估值逻辑的崩塌迅速波及一级市场。Cembalest 分享了一个切身经历：他每两年撰写一份关于私募股权（PE）、私募信贷（Private Credit）和对冲基金的另类投资报告。去年 11 月，他将私信贷章节初稿发给合作伙伴黑石集团（Blackstone）的一位高管审阅。当时对方严厉批评 Cembalest，指责他严重低估了黑石资产组合中庞大软件敞口的价值主张。然而仅仅过了几个月，到了次年 2 月，市场情绪彻底逆转，软件资产从昔日的金牌卖点沦为了大家唯恐避之不及的话题，曾经坚不可摧的 Salesforce 等老牌软件帝国正在遭遇前所未有的生存质疑。 - **续约常态被打破**：传统 SaaS 每年 3% 的无脑涨价契约不再是板上钉钉的“送分题”（layups）。 - **叙事反转速度极快**：去年 11 月头部机构仍在力推软件贷款资产的高确定性，短短一个季度后，市场便全面转向对传统软件生命周期的深刻怀疑。 ### [52:49](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3169s) ASIC 崛起、Nvidia 估值挤压与杠杆 ETF 引发的市场异动 根据摩根大通股票研究数据，各大巨头自研 ASIC（专用集成电路）的单体经济效益正在急剧改善，包括 Google 的 TPU、AWS 的 Trainium、微软的 Maia 以及 Meta 的 MTIA。采用自研芯片可为超大规模云厂商降低 30% 至 40% 的总体拥有成本（TCO）。Batnick 提出疑问，这是否正是英伟达（Nvidia）股价陷入滞胀、远期市盈率（Forward P/E）被压缩至 18 到 20 倍低位的主因？Cembalest 赞同这一逻辑，指出过去市场盲目迷信英伟达 GPU 的不可替代性，认为云厂商只能被迫接受英伟达高达 70% 的毛利率；然而如今事实证明，各大厂商的自研算力迁移战略已全面提速。尽管这些自研加速器目前主要供内部工作负载使用、缺乏英伟达强大的销售与软硬件生态支持，但由于他们本身就是英伟达最大的客户，哪怕只是部分替代自身原有的采购配额，对英伟达的长期增速而言也是巨大打击。Brown 提及亚马逊 CEO Andy Jassy 的言论，若将内部 Trainium 芯片业务按半导体公司独立估值，其年化营收体量足以跻身全球最大公司之列。 与此同时，微观层面的投机与杠杆正在严重扭曲市场价格发现机制。Batnick 指出，目前双倍杠杆 ETF（Double-leveraged ETFs）的名义敞口规模已激增至约 5,000 亿美元。Cembalest 引用研究报告中针对衍生品交易对日内市值波动贡献的图表解释道，杠杆 ETF 发行商为维持日常对冲中性，必须严格执行“追涨杀跌”的动态再平衡机制——上涨时追加买入，下跌时被动抛售，并确保每天收盘前抹平风险头寸。这种被动资金机制导致美股交易日在下午 2:30 至 4:00 之间频现极端的无序剧烈震荡，极大消耗了普通投资者的神经。一旦相关个股从高点回落进入 10% 到 15% 的震荡横盘期，内嵌杠杆的波动率衰减（Volatility Drag）将给产品净值带来毁灭性打击，这种疯狂的投机潮也才会真正降温退场。 ### [57:26](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3446s) 韩国散户杠杆激增与科技巨头资本开支融资警讯 在讨论全球资本流动与狂热情绪时，Michael Cembalest 指出图表中最引人注目的警示信号莫过于韩国综合股价指数（KOSPI）的保证金贷款（Margin Loans）激增。回顾 1997 年亚洲金融危机与 2008 年全球金融危机，Cembalest 直言韩国散户历来是“最可怕的市场逆向指标”——一旦这些散户开始蜂拥入场并全力加杠杆，通常意味着行情已经走到了非常后期的阶段。尽管目前散户疯狂追逐三星（Samsung）和 SK 海力士（SK Hynix）具有基本面支撑（两家公司的增长态势极其惊人），但他们仍像是“挂在列车最末端的守车”。未来六个月内，市场急需超大规模云厂商（Hyperscalers）预期现金流与 AI 落地采用率实现强劲反弹，才能消化当下的高估值与狂热。 围绕科技巨头的融资动作，主持人 Josh Brown 与 Michael Batnick 探讨了市场分歧。对于亚马逊（Amazon）发售 300 亿美元债券，Cembalest 并不感到担忧，认为其债务与现金流比率依然健康、规模尚在承受范围内；然而 Alphabet 在进行了长达十年的股票回购后突然转为发售股票募资，性质就截然不同。这种通过股权融资来支撑资本支出（Capex）的行为，直接暴露出其内部产生的自由现金流已经无法单纯通过“开支票”来全额覆盖巨额开支。 - **韩国散户保证金贷款**：历史经验表明散户大规模加杠杆进场往往是周期极度见顶的危险信号。 - **内存双雄驱动**：三星电子与 SK 海力士的强劲业绩吸引了韩国本土资金不计后果地涌入。 - **债务 vs 股权融资**：巨头发行债券尚属正常资本运作，但依靠增发股票填补资本开支则引发对其现金流生成能力的担忧。 --- ### [58:58](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3538s) SpaceX 上市锁定期机制与纳入指数之争 话题转向 SpaceX 的上市与交易机制。主持人探讨了传统 IPO 模式中高度套路化且效果糟糕的 6 个月锁定期问题——通常解禁期一到股价便断崖式下跌。SpaceX 采取了分批错峰的阶梯式锁定期（Staggered Lockup），并设定了不同触发价格。虽然仅释放 3% 的流通股引发了一定微词，但这一创新设计依然被视作有益的尝试。Cembalest 指出，作为 IPO 的替代方案，直接配售（Direct Placements）的实际效果仍待观察；预计到今年年底，SpaceX 的自由流通股比例将从最初的 3% 左右攀升至约 50%。 围绕 SpaceX 的市场情绪充满了戏剧性。主持人甚至调侃英国《金融时报》旗下 FT Alphaville 给出的一篇将 SpaceX 12 个月目标价定为“无穷大”（Infinity）的极度讽刺性评级，引得现场一片欢笑与无奈。此外，早先许多专家曾发出严重警告，宣称 SpaceX 的上市将引发资金抽血、搞垮大盘指数，但事实证明这一悲观预测并未发生。大盘纳斯达克 100 指数 ETF（QQQ）表现稳定，标普道琼斯指数公司（S&P）也顶住了压力，坚守了成分股必须具备盈利能力的硬性标准，未向亏损企业破例；相比之下，纳斯达克因其交易所属性已迅速将其纳入。 ```mermaid flowchart TD accTitle: SpaceX Listing and Index Inclusion Flow accDescr: 展现 SpaceX 的股权解禁机制与标普、纳斯达克两大指数机构不同的纳入策略 subgraph ListingMechanics["SpaceX 发行与解禁机制"] A["初始上市发行<br>仅售 3% 极低流通盘"] --> B["阶梯式分批锁定期<br>结合特定股价触发点"] B --> C["年内流通比例提升<br>预计年底 Free Float 达 ~50%"] end subgraph IndexDecisions["主流指数纳入策略分歧"] D{"指数准入考量"} D -->|"纳斯达克 (NASDAQ)"| E["直接纳入基准指数<br>(依托本所挂牌上市)"] D -->|"标普指数 (S&P)"| F["坚守严格门槛<br>(必须满足持续盈利标准)"] end ListingMechanics -.-> IndexDecisions ``` - **锁定期创新**：避免传统 IPO 到期集中抛售的踩踏，通过阶梯释放平抑波动。 - **指数分工妥协**：标普坚持盈利标准避免指数“垃圾化”，纳斯达克先行吸纳，形成了市场折中平衡。 --- ### [01:01:45](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3705s) 牛市健康体检：订单库存比与中期选举噪音 对于当前美股牛市的健康程度，主持人和嘉宾从内部指标展开了细致评估。市场上行呈现出不少积极特征：区域银行、运输板块、工业股、小盘与微盘股，乃至以花旗集团（Citigroup）为代表的华尔街大行均展现出强劲动能，零售销售数据与银行贷款增长亦大幅超出预期。尽管超大规模云厂商和半导体估值存在剧烈分化，市场广度（Market Breadth）也不算完美，但经济底色依然具备韧性。 Cembalest 特别强调，在他所跟踪的数百个宏观数据中，与美股走势相关性最强的指标是**制造业采购经理人指数（PMI）中的“新订单与库存之比”**（Ratio of New Orders to Inventories）。标普 500 本质上是一个生产型指数而非纯消费指数，因此新制造订单相较于库存积压的强弱，是预测企业盈利和股市走势的核心先导指标，目前该数据依然稳健。虽然信用卡及汽车贷款违约率呈现上升趋势，暴露出低端消费者的疲软，但这不足以动摇标普大盘的盈利支柱。谈及下半年的中期选举，Cembalest 认为大选对股市通常只是充满杂音的短期干扰，正如国情咨文一样，极少对长期投资组合产生实质性的政策颠覆。 - **核心风向标**：制造业 PMI“新订单 / 库存”比率是研判标普 500 最具前瞻性的指标。 - **分化现实**：汽车和信用卡违约率走高反映大众消费压力，但企业生产与大行贷款仍维持景气。 - **政治噪音免疫**：除非中期选举导致两党对立并彻底瘫痪关键财政政策，否则难以改变大盘主趋势。 --- ### [01:04:54](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3894s) AI 独角兽上市前景与业务不确定性排序 针对市场高度关注的头部 AI 独角兽（OpenAI 与 Anthropic）能否在年内顺利上市，主持人担忧若上市延期，是否会冲击整个 AI 生态链供应商的业绩预期。Cembalest 对流动性与承接力持乐观态度：目前标普 500 企业的回购力度依然充沛，年内并购交易（M&A）中现金支付比例高达约 70%，加之若新股同样采取低流通盘（如仅发售 3%）的策略，市场完全有充裕的资金予以消化。 然而在企业底层业务的确定性上，Cembalest 给出了明确的风险梯队划分。如果剥离上市估值溢价与短期市场杂音，从核心商业模式的底层不确定性（Business Uncertainty）来看：**OpenAI 面临的最大，其次是 Anthropic，而 SpaceX 的业务确定性最高、不确定性最小**。主持人对此观点表示认同，并直言公开市场届时很可能对 OpenAI 的高额亏损与商业化路径表现出严苛甚至排斥的态度。 - **并购与流动性**：年内 70% 的并购由纯现金支付，显示流动性池并未枯竭。 - **业务不确定性排行（由高到低）**： 1. **OpenAI**（不确定性最高：巨额算力投入、高烧钱率、激烈的商业模式探索） 2. **Anthropic**（不确定性居中：侧重企业级安全与模型研发，面临巨头竞合） 3. **SpaceX**（不确定性最低：星链商业化清晰、发射订单壁垒极高） ### [01:06:19](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=3979s) 商业不确定性与早期网络效应的脆弱性 讨论延续了对前沿 AI 巨头的对比评估。Michael Cembalest 指出，在他看来，OpenAI 面临的商业不确定性远高于竞争对手。造成这种脆弱性的核心原因在于其收入结构的高度集中：OpenAI 并没有庞大的广告收入支撑，主要依赖消费者订阅模式，企业端业务相对较弱。这种严重偏向 C 端消费者的模式极易受到市场情绪波动的冲击。 Josh Brown 赞同这一判断，并将 OpenAI 当前的处境比作早期互联网时代的 AOL（美国在线）——它是几乎所有人最先接触和使用的产品，但随着生态成熟，用户绝不会止步于此。虽然先发优势带来的网络效应十分强大，但 Anthropic 等竞争对手的迅速崛起已经证明，市场格局和用户偏好的转变可以发生得极其迅猛。 - **商业模式短板**：OpenAI 缺乏多元化收入来源，高度依赖个人消费端订阅费，相较于深度绑定 B 端和基础设施的企业，抗情绪周期能力较弱。 - **“AOL 效应”警示**：早期爆款产品极易在技术和生态更迭中被后续更成熟或更细分的竞品分流。 --- ### [01:07:10](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=4030s) Project Glasswing、“Patchmageddon”与 AI 带来的网络安全狂欢 Michael Cembalest 透露了他当下的核心夏季研究项目：摩根大通（JPMorgan）作为“Project Glasswing”的成员单位之一（该联盟由 20 到 30 家获准提前接入 Mythos 等尖端网络安全模型的核心企业组成），首席运营官（COO）要求他与网络安全团队合作撰写一篇重磅报告，梳理他们从中获得的洞察及未来趋势。该报告被形象地命名为《Patchmageddon》（补丁末日）。 ```mermaid flowchart TD accTitle: AI-Driven Vulnerability Discovery and Cybersecurity Gap accDescr: Diagram illustrating how AI models uncover obscure vulnerabilities faster than humans can patch them, leading to security risks. subgraph AI_Discovery["AI 深度探索"] A["尖端模型 (如 Mythos)"] --> B["超人类遍历与排列组合"] B --> C["发现极其隐蔽的系统漏洞"] end subgraph Security_Crisis["防御端失衡 (Patchmageddon)"] C --> D["漏洞确认速度爆发式上升"] E["补丁修复与部署节奏滞后"] D -. 产生巨大缺口.-> E end subgraph Attack_Surface["潜在受威胁领域"] D --> F["国家背景与黑客组织活动"] D --> G["可编程逻辑控制器 (PLC) 及物理基建"] D --> H["各体量上市企业遭受突发攻破"] end ``` 这种安全威胁的逻辑与 AI 在药物研发领域的突破极其相似：就像 AI 能够通过穷尽人类思维从未尝试过的分子排列组合，从而成功找到对抗耐甲氧西林金黄色葡萄球菌（MRSA）的全新抗生素一样，AI 同样能用极其反直觉的异构操作路径（例如某种特定的登录顺序、辅以特定异常操作及按键敲击）绕过防御，触发企业级系统的严重失陷。 Josh Brown 感慨，这种神经网络的运作逻辑就像“降临在我们身边的外星生命”，它并不按照人类的思维方式推演，即便给出的答案符合人类预期，其黑盒机制也让构建者无法逆向解析推导过程。Cembalest 进一步以银行业举例：两三年前各大银行尝试用神经网络实时审批或拒绝借记卡与信用卡刷卡交易，但由于无法向监管机构解释为何某笔交易被拒，银行被迫在顶层叠加各种翻译工具以将机器决策转化为人类语言，但效果并不理想。这种不可解释性意味着系统可能正确运行 100 次，却在第 101 次产生灾难性后果。Cembalest 警告，尽管这不一定会直接拖垮整体大盘，但各体量的上市公司都必须系好安全带，准备迎接频发的网络入侵头条风暴。 - **Project Glasswing 与报告核心**：摩根大通深度参与的高级别安全研究项目，聚焦高阶 AI 挖掘未知系统缺陷的能力，揭示了“漏洞激增 vs. 补丁滞后”的失衡危局。 - **物理基建风险**：工业控制系统中的可编程逻辑控制器（PLC）等老旧物理基础设施面临极高暴露风险。 - **黑盒不可解释性**：神经网络在金融反欺诈等领域的实际应用受制于监管要求，因缺乏因果归因链条而难以实现稳健合规。 --- ### [01:10:36](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=4236s) 智力消耗、跨代赞誉与撰写研报的职业教训 面对技术与宏观交织的复杂世界，主持人好奇 Cembalest 是否正沉浸在破解这些技术谜题的极大乐趣中。Cembalest 却苦笑否认：“我都 64 岁了，这并不好玩。为了维持同等的产出，我的大脑单位负荷要大得多。”主持人则高度称赞 Cembalest 的研究地位，直言每次他发布研报就像年轻一代在翘首以盼 Frank Ocean 发布全新专辑一样，其跨周期的穿透力和专业度为市场提供了巨大的洞察。 谈及面对公众反馈时的心态，Cembalest 分享了自己职业生涯中历经的一次深刻教训。作为他每年耗时三四个月打磨的旗舰级年度能源报告，今年一期的主题为《Fighting Words》（争论之言），封面借鉴了经典西部电影《黄金三镖客》（The Good, the Bad and the Ugly）中三方对峙的决斗场景，暗喻核能、可再生能源与化石燃料三大阵营在社会与政府政策内部的激烈角力。 在起草该报告的导言时，Cembalest 匿名引用了一位曾与其发生过极度激烈争吵的客户的真实互动，将其作为能源领域观点撕裂的隐喻。尽管未指名道姓，但由于细节过于具体，该客户一眼便认出自己，并感到极为愤怒，导致 Cembalest 的直属上级、摩根大通资产与财富管理部 CEO Mary Erdos 亲自致电介入。经历此次风波后，Cembalest 深刻意识到研报撰写中的底线不能逾越。作为最终的建设性解决方案，他在职业生涯中首次破例：在来年的年度能源报告中，他将允许该客户指定一位政策能源专家，在报告中撰写独立、未经删改的一整节内容，公开且正面地陈述与 Cembalest 相左的立场与依据。 - **高负荷的研究现实**：面对前沿技术的跨学科冲击，老牌宏观策略师需要承受更高的认知负荷与智力支出。 - **《Fighting Words》封面寓意**：借用《黄金三镖客》的三枪决斗隐喻核能、新能源与传统化石能源之间的零和博弈与政策撕裂。 - **化解争端的开放机制**：在下一期年度能源报告中引入反对派专家的“原汁原味”客座反驳专栏，将客户摩擦转化为极具学术与行业价值的公开思辨。 ### [01:15:15](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=4515s) 追踪 Michael Cembalest 的研究与 250 期演播室庆祝 Josh Brown 与 Michael Batnick 在节目的收尾环节向观众推荐了如何获取 Michael Cembalest（来自摩根大通资产管理）的高质量宏观研报与观点。主持人们提到，关注者除了可以在 LinkedIn 上追踪他的最新动向外，Cembalest 每期发布《Eye on the Market》报告时，还会配套推出由他在家中录制的精简版视频播客（video digest）。Batnick 也特意表达了对该视频系列的高度喜爱。Cembalest 更是幽默地补充了一句，观众如果感兴趣，还可以在 Instagram 上关注他的“钓鱼探险日常”。 正当几位主持人准备顺势结束录制时，演播室现场突然出现了意料之外的庆祝环节。现场工作人员端着庆祝道具登场，引起主持人们一阵紧张与玩笑，纷纷喊着“千万别是鞭炮”（"No firecrackers."），甚至笑称现场负责人 Rob 必须承担起安全职责。随后现场点燃了庆祝冷烟火/火花棒，全场齐声高呼“Happy 250 everybody!”，庆祝《The Compound and Friends》迎来具有里程碑意义的第 250 期。 面对演播室内热烈且略带失控的搞笑氛围，Cembalest 忍不住感叹“这真是一档独一无二的节目”（"This is a unique program"）。大家互相开着玩笑，主持人们调侃制作人员 Nicole 随后在脚注里的评价，甚至抛出文学梗“Holden Caulfield, here we come”。最后，Josh Brown 与 Michael Batnick 再次向嘉宾 Michael Cembalest 致谢，同时向所有收看与收听播客的观众表达感谢，并在现场热烈的掌声与背景音乐中为本期 250 期特别节目画上了圆满的句号。 ## Notable quotes > [01:08](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=68s) 当犯错的成本极高时，采用的门槛就必须定得非常高。 — Michael Cembalest > [09:46](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=586s) 在历史上所有的技术革命中，这一场是最难理解各个细枝末节的。 — Michael Cembalest > [16:09](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=969s) 我不需要花钱请分析师告诉我 8 倍市盈率的股票便宜；我需要的是有人告诉我，为什么 40 倍市盈率的股票反而是便宜的，而 8 倍市盈率的却可能极其昂贵。 — Michael Cembalest > [44:08](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2648s) 如果你对这些话题抱有弥赛亚般的狂热信仰，或者无法容忍任何异见，请勿阅读。 — Michael Cembalest > [38:07](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2287s) 标普 500 本质上是一个‘生产指数’，而宏观经济则是一个‘消费指数’。 — Michael Cembalest > [39:22](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=2362s) 上游组件和光学器件供应商往往是整列火车里‘最后拿到备忘录’的车厢，车头的引擎早就撞毁跌下悬崖了，坐在末节车厢的供应商看着眼前堆积如山的订单簿，还误以为形势大好。 — Downtown Josh Brown > [01:10:36](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=4236s) 我都 64 岁了，这并不好玩。为了维持同等的产出，我的大脑单位负荷要大得多。 — Michael Cembalest > [01:15:15](https://www.youtube.com/watch?v=-mRmA5ZUhmE&t=4515s) 这真是一档独一无二的节目。 — Michael Cembalest ## Mentioned Michael Batnick, Downtown Josh Brown, Michael Cembalest, J.P. Morgan, Eye on the Market, Federated Hermes, Bridgewater Associates, OpenAI, ChatGPT, Anthropic, Claude, Google, Gemini, 智谱 GLM-4 / 5, Open Router, Nvidia, ASML, Apple, Alibaba, Kevin Warsh, Brad Setser, Euroclear, Simpson-Bowles Commission, Bloom Energy, Micron Technology, TSMC, Cisco, Gary Brinson, Michael Burry, Blackstone, Salesforce, Amazon, Andy Jassy, Samsung Electronics, SK Hynix, SpaceX, FT Alphaville, Citigroup, Project Glasswing, Mary Erdos ## Video description On episode 250 of The Compound and Friends, ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Michael Batnick⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Downtown Josh Brown⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ are joined by ⁠Michael Cembalest⁠ to discuss: the strength of the U.S. dollar, China’s challenge to American economic dominance, foreign demand for Treasuries, the federal debt reckoning, the rapidly changing AI investment landscape, Nvidia’s growing competition, semiconductor cyclicality, and more! This episode is sponsored by Federated Hermes. Explore their full ETF lineup at ⁠https://federatedhermes.com/us ▶00:00 - Cold Open ▶13:06 - Michael Cembalest Joins the Show ▶16:09 - Why Betting Against America Hasn’t Worked ▶23:56 - The Truth About Central Bank Gold Buying ▶25:22 - Is the U.S. Dollar Losing Its Dominance? ▶29:00 - Is China Really Dumping U.S. Treasuries? ▶32:44 - America’s Federal Debt Time Bomb ▶39:20 - Are Memory and Semiconductor Stocks in a Bubble? ▶46:09 - AI: The Most Complex Market Catalyst in 40 Years ▶51:41 - How AI Is Disrupting Software and Private Markets ▶58:57 - SpaceX, IPO Lockups, and Index Inclusion ▶01:01:41 - How Healthy Is the Bull Market? ▶01:03:52 - Will the Midterms Move the Stock Market? ▶01:04:50 - OpenAI vs. Anthropic vs. SpaceX ▶01:07:10 - AI’s Growing Cybersecurity Threat 👕 Check out The Compound shop: https://www.idontshop.com 🎙️Listen to our podcasts: TCAF: https://podcasts.thecompoundnews.com/show/TCAF/ Animal Spirits: https://podcasts.thecompoundnews.com/show/animalspirits/ Ask The Compound: https://podcasts.thecompoundnews.com/show/ATC/ The Compound Newsletter: https://www.thecompoundnews.com/subscribe Linkedin: https://www.linkedin.com/company/the-compound-media/ Instagram: https://instagram.com/thecompoundnews Twitter: https://twitter.com/thecompoundnews Tik Tok: https://www.tiktok.com/@thecompoundnews Investing involves the risk of loss. This podcast is for informational purposes only and should not be or regarded as personalized investment advice or relied upon for investment decisions. Josh Brown and Michael Batnick are employees of Ritholtz Wealth Management and may maintain positions in the securities discussed in this video. All opinions expressed by them are solely their own opinion and do not reflect the opinion of Ritholtz Wealth Management. See our disclosures here: https://ritholtzwealth.com/podcast-youtube-disclosures/ The Compound Media, Incorporated, an affiliate of Ritholtz Wealth Management, receives payment from various entities for advertisements in affiliated podcasts, blogs and emails. Inclusion of such advertisements does not constitute or imply endorsement, sponsorship or recommendation thereof, or any affiliation therewith, by the Content Creator or by Ritholtz Wealth Management or any of its employees. For additional advertisement disclaimers see here https://ritholtzwealth.com/advertising-disclaimers. #thecompound #investing #stockmarket #stocks [Source](https://www.youtube.com/watch?v=-mRmA5ZUhmE)

[1:08] Michael Cembalest: 当犯错的成本极高时，采用的门槛就必须定得非常高。 [9:46] Michael Cembalest: 在历史上所有的技术革命中，这一场是最难理解各个细枝末节的。 [16:09] Michael Cembalest: 我不需要花钱请分析师告诉我 8 倍市盈率的股票便宜；我需要的是有人告诉我，为什么 40 倍市盈率的股票反而是便宜的，而 8 倍市盈率的却可能极其昂贵。 [44:08] Michael Cembalest: 如果你对这些话题抱有弥赛亚般的狂热信仰，或者无法容忍任何异见，请勿阅读。 [38:07] Michael Cembalest: 标普 500 本质上是一个‘生产指数’，而宏观经济则是一个‘消费指数’。 [39:22] Downtown Josh Brown: 上游组件和光学器件供应商往往是整列火车里‘最后拿到备忘录’的车厢，车头的引擎早就撞毁跌下悬崖了，坐在末节车厢的供应商看着眼前堆积如山的订单簿，还误以为形势大好。 [70:36] Michael Cembalest: 我都 64 岁了，这并不好玩。为了维持同等的产出，我的大脑单位负荷要大得多。 [75:15] Michael Cembalest: 这真是一档独一无二的节目。

[23:56] Michael Cembalest: Gold prices have gone up... The actual allocated troy ounces of gold that these central banks own has fallen in valuation terms... 95% of the reason why gold as a share of foreign exchange reserves has gone up is price, not because central banks are chasing it and hoarding. [32:44] Michael Cembalest: In 2031 or 2032 you reach this crossover point where 100% of federal tax revenue has to be used to pay interest on the debt and entitlements with nothing left for anything else. That is a scary day. [39:20] Michael Cembalest: The hyperscalers are starting to roll over even though the semiconductor index is zooming... If the big four hyperscalers plus Oracle can't convince the market that they're earning a good return on a trillion-five of capital spending outlays, those infrastructure stocks are at risk. [51:56] Michael Batnick: Once you give people these tools, the only thing they can think about is what else can this do? [52:10] Michael Batnick: For the last few years we were all told that Nvidia GPUs were untouchable... That was wrong. [58:30] Michael Cembalest: Korean retail is basically the scariest market indicator because by the time those guys start to pile into stuff, you know you're really, really, really late in the game. [58:57] Michael Batnick: Somebody has to have some profitability standard, or else you end up with a Russell 2000—which is a pile of companies that don't make money. [64:50] Michael Cembalest: I think if I had to rank the business uncertainty, I would say OpenAI the most, then Anthropic, then SpaceX. I don't think the market's going to like OpenAI. [67:10] Michael Cembalest: The pace of confirmed vulnerabilities is going up like this, and the pace of patching is down here... State-sponsored and non-state-sponsored bad actors are going to have a field day for some period of time using these models to discover bizarre hidden vulnerabilities that nobody knew existed. [73:36] Michael Cembalest: In next year's energy paper, I'm going to allow that person to select the policy energy expert of their choice to write an unedited, unexpurgated section in the energy paper on why they disagree with me and on what.

[0:00] Michael and I are talking his birthdays [0:01] in May. He used to be able to fish in [0:03] shorts. We May is gone. [0:04] >> Yeah. May is very May is very cold now. [0:07] >> Oh. In New York, the thing that people [0:08] don't realize is September, October is [0:10] better than April and May [0:11] >> weatherwise. [0:12] >> Yeah. [0:13] >> And pe people forget like people get [0:15] excited. Oh, it's spring. Spring is not [0:17] great in New York. [0:18] >> No, it's not. [0:18] >> Fall is wonderful in New York. Um, what [0:21] is the Latin term that you use to [0:23] describe the eagle octopus? [0:25] >> The aquilisf. [0:26] >> Aquil. So Aquilla is lat Latin for [0:29] eagle. [0:30] >> Yes. [0:30] >> Okay. [0:31] >> And sephalopod as in [0:34] >> which is the the prefix used in in Latin [0:36] and in biology for [0:37] >> it's a little bit of a disturbing image [0:39] but I but it's uh it catches your [0:43] attention for sure. Okay. [0:44] >> Yeah. I mean what you know GP you know [0:46] the GPT is amazing now in terms of you [0:48] you can give it very specific guidance [0:50] in terms of what you wanted to draw. [0:52] >> I was going to ask if this was AI [0:53] generated or you had an artist inside of [0:54] JPM. We, you know, we used to use [0:57] external artists, but it's just it's so [0:58] much easier now and uh even compared to [1:01] a year ago, the the degree to which you [1:04] can continue to layer on additional [1:06] instructions and have it create whatever [1:08] image you want. [1:08] >> How much does that frustrate you when [1:10] people are talking about or complaining [1:11] about what AI can't do when it's [1:14] evolving so so fast? Like we're looking [1:16] at this in a vacuum as if it's a [1:18] finished product sometimes when people [1:19] are talking. [1:20] >> Yeah. Look, I think I I like the fact [1:22] that when people do empirical analyses [1:25] that track how it's doing. Um, you know, [1:28] it's still [1:30] it's getting better at certain things, [1:31] but you know, think about it this way. [1:33] When the cost of being wrong is very [1:35] high, I think the hurdle should be very [1:37] high. Uh, you know, there were there was [1:40] a time a few years ago when people felt [1:41] like radiologists wouldn't be needed [1:43] anymore. Um, if you if you are at risk [1:45] of breast cancer, you still want a [1:47] doctor plus AI instead of AI by itself. [1:50] Um, and so it's it's still not ready for [1:52] prime time on certain interpretive [1:54] things in medicine, but it is getting [1:56] better at at at things that have low [1:59] value, like making drawings of [2:00] >> I actually eagle octtopi. [2:02] >> I actually like that it's not perfect [2:05] >> because I like to yell at it. I get a [2:08] lot of satisfa I can't say the things [2:10] that I can say to my uh claude to other [2:13] people. So I like to just put it all [2:15] there and I find it to be [2:17] >> that's poor AI. Well, I'm I'm [2:19] >> definitely going to get kicked off of [2:20] this this Claude. [2:22] >> I'm not easy I'm not easily kind of [2:23] surprised, but I am very surprised about [2:26] the number of people that use it as a [2:28] psychiatrist or a counselor or for for [2:31] marriage advice and things like that. [2:33] >> My wife my wife is talking to it about [2:35] nutrition all day and used to pay a [2:39] nutritionist like go every week. Here's [2:42] what I ate. Here's what I should have [2:43] ate. here's the choices that I made at [2:46] this like Italian restaurant. That makes [2:48] sense. That's mostly factual. [2:49] >> She's she much prefers talking to Chachi [2:52] BT as an as it's not that it's her [2:55] nutritionist. It's tracking the minutia [2:58] of what she eats and doesn't eat each [3:00] day. And she just likes that experience [3:02] better because it's whenever she feels [3:05] like it. [3:06] >> It's it's not I have an appointment. [3:07] It's like anytime I feel like talking [3:09] about what I'm what I'm doing, I could [3:11] talk to it. The I mean these things are [3:12] slavishly trained to be so obsequious to [3:15] you that you have to be careful exactly [3:18] what you do rely on. Although I do have [3:20] this, you know, like a lot of us, I have [3:22] this fancy Manhattan internist and [3:24] whenever I call him and I'm describing [3:25] my symptoms, I can hear him typing them [3:28] into the computer, [laughter] [3:30] giving me a readout from Google [3:32] overview. [3:33] >> Well, they ask you for relationship [3:34] advice. That seems a bit I know people [3:36] are doing it. Seems a a bit a bit [3:38] strange. [3:38] >> Uh yeah. Yeah. I mean, maybe that's one [3:41] of the reasons that you're having a [3:42] problem with your relationship to begin [3:44] with is that you kind of are trusting in [3:46] a in a computer, [3:47] >> but I I I I think it's the convenience [3:50] of always on more so than it's I trust [3:52] the result from this more than I trust a [3:54] person. And so, professional services [3:57] organizations are going to have to find [3:59] a way to take their own internal [4:01] knowledge base and put them in a setting [4:04] where somebody who wants to access the [4:07] way they would have answered a question [4:09] >> Yeah. can can be uh 24/7 always on and [4:12] it's not really replacements and [4:14] extension. [clears throat] [4:15] >> Yeah. One of the things that's amazing I [4:16] mean I'm sure we'll get into this when [4:17] we get into the meat of the discussion [4:19] is [4:22] free openweight models trained on a [4:25] company's own data. [4:26] >> Yeah. [4:27] >> Can now outperform the frontier labs. [4:30] >> Yeah. [4:30] >> Which are much more expensive. And so [4:32] that's kind of a huge deal. Bridgewwater [4:34] published something recently showing [4:36] that an offthe-shelf completely free [4:39] model um had better results at maybe 5% [4:43] of the cost of the front of OpenAI. I [4:46] read what they said. So, they are [4:48] feeding all of these economic uh data [4:51] releases and news articles and research [4:55] into this thing and they're trying to [4:57] see how valuable uh how valuable [5:02] each different item is and they're [5:05] trying to like weight the response of [5:07] the AI to these things. It's it's it's [5:10] [clears throat] designed to collate [5:13] all the information that exists [5:14] overnight and create a reading package [5:17] for the portfolio managers and the [5:19] >> pay attention to this, don't pay [5:20] attention to that. [5:21] >> And so, you know, they're they're using [5:23] some qualitative judgments, but what [5:25] they're finding is that the free model [5:27] trained on their own data does a better [5:29] job than some of the frontier labs, [5:30] which are much more expensive. I mean, [5:32] this is something that we're going to [5:33] have to watch as we start thinking about [5:35] where we are. Um just over the last 48 [5:39] hours or so some data is coming out of [5:41] open router showing that almost 50% of [5:45] all of the tokens from US companies are [5:48] being directed to Chinese models rather [5:50] than the US frontier [5:51] >> more efficient. [5:52] >> So well less token per use. [5:55] >> Yes. Yeah. Like if you if you if you if [5:57] you're trying to figure out a cost per [6:00] performance [6:02] th those models are cheaper. And I don't [6:03] think 12 months ago or even 6 months ago [6:05] people that was going to happen quite so [6:07] quickly. [6:08] >> What is [6:10] what is the reason for the for the [6:12] superior uh efficiency of the Chinese [6:15] model? They're just distilling the US [6:18] model but using less inputs. [6:20] >> They have no MSJ. [6:22] >> No. [6:23] [laughter] [6:25] >> So [6:26] >> like why why would why would somebody [6:28] route something to a Chinese model and [6:29] why would it be more efficient to do [6:31] that? Well, I mean the the the bigger [6:34] question is why is China spending [6:38] billions of dollars [6:42] creating models that they're giving away [6:43] for free, right? I don't think there's [6:45] any question that the performance of [6:48] those models, you know, every 3 to six [6:50] months closes the gap with the US ones. [6:52] There's all sorts of data from [6:54] artificial intelligence and ELO scores [6:56] and all of the different benchmark GPQA [6:58] diamond scores. I mean there's a whole [7:00] cottage industry of assessing reasoning, [7:03] math, coding, language capabilities of [7:06] these models and um the uh GLM 5.2 which [7:10] is the new Chinese one is a hair's [7:13] breath away from GPT 5.5 noticeable for [7:16] the average user. For the average user [7:17] it's perfectly fine. [7:18] >> Perfectly fine. Yeah. And so the qu the [7:21] question is why are the Chinese [7:24] companies putting so much money into [7:27] things they're giving away for free [7:29] >> thoughts? [7:30] >> Well, there's a lot of debate about this [7:32] >> people inputting data there. [7:33] >> There's I there's a lot of debate about [7:35] this. I I I can't escape the fact that [7:38] the most valuable [7:41] companies in the US universe are either [7:44] the Frontier Labs themselves or the [7:47] companies that sell hardware and [7:49] equipment to those frontier labs. And if [7:51] you're China, that is a it if you can [7:54] stick a pin in that balloon, that [7:56] achieves probably more of your long-term [7:58] objectives than a lot of other things [8:00] you might be doing like industrial. It's [8:01] a competit. They're they're rendering [8:03] the the advantage that we have less [8:06] valuable. [8:08] >> Oh my god. China cap keeps growing. So I [8:11] don't [8:12] >> if China could wake up one day and and [8:14] and figure out a way to make the large [8:18] the mega cap companies worth a third of [8:20] what they are right now. They'd do it in [8:21] a second and they'd spend a lot of money [8:23] doing it. And so I I now there you can [8:25] also make some arguments that they're [8:27] trying to create an ecosystem within [8:29] China that relies on their own compute [8:31] and things like that. And it's probably [8:33] a little bit of mix of everything. But [8:35] you know it is it is remarkable to see [8:38] these companies creating and providing [8:42] these free tools that you that you can [8:45] now run you know on your own devices. [8:48] You don't necessarily none of your data [8:50] has to be routed to Chinese servers. you [8:52] can run the whole thing in house. You [8:54] can even run it offline, you know, if [8:56] you've got a powerful enough Nvidia chip [8:58] on your on a laptop. [9:00] >> So, um it's it's there's I can imagine [9:04] at some point that the Department of [9:06] Commerce is going to have to start [9:07] thinking about um the impact of those [9:10] Chinese models and and do they want them [9:12] proliferating in the same way that [9:14] Europe is trying to figure out whether [9:15] or not they should put a cap on the [9:17] importation of Chinese EVs, which are [9:19] decimating the European auto industry. I [9:21] was going to say the the Europeans are [9:23] still trying to agree on a definition [9:24] for AI. Uh I remember from the social [9:27] media age the question was why are they [9:29] giving away Facebook? Why like why is [9:32] why are all these new products being [9:34] created and given away for free? And the [9:36] answer is if you're not paying you're [9:39] not the customer. You might be the [9:41] product. [9:41] >> Yeah. [9:42] >> They're not giving away anything for [9:44] free. They're serving you to [9:45] advertisers. like [9:46] >> right [9:47] >> there's probably something to that where [9:50] it's more valuable it's more of a [9:52] strategic priority in China to have us [9:55] and other international users putting [9:59] information into these right [10:01] >> programs in as an answer to one of the [10:03] other things you mentioned I I I look I [10:05] talked to a lot of people this is way [10:06] above my pay grade I have to say [10:08] [clears throat] of all the technological [10:09] revolutions that have ever happened [10:12] this one's the hardest to kind of [10:14] understand the bits and pieces of I mean [10:16] compare this to the fiber buildout and [10:18] the internet and B2B software. I mean [10:20] those are pretty simple compared to [10:21] this. But the people that I talk to that [10:24] live in this world say that you know a [10:28] small component of Chinese success has [10:30] come from training their models on US [10:32] models. Yeah. [10:33] >> Um but there's a lot of independent [10:35] engineering that they're doing on our [10:36] end as well. We shouldn't we shouldn't [10:38] underestimate the degree to which the [10:42] controls that have been put in place by [10:44] the US have forged a an ecosystem in [10:47] China that is determined to figure out, [10:50] >> pardon [clears throat] me, it's its own [10:51] way of building these things out on its [10:53] own. They now have an alternative to the [10:55] ASML lithography machines. They're not [10:57] quite as good, but they're getting [10:58] there. [10:58] >> Yeah, I think I think it's pretty [11:00] obvious that you're right about that. [11:01] And you know, Apple's got a Apple's got [11:04] a massive business in China. Their US [11:08] partner for AI is going to be Gemini. [11:10] It'll be Google. That's not going to [11:12] work in China. So, they have a deal with [11:14] Alibaba. And they're going to use models [11:16] that are coming, I guess, from Alibaba. [11:19] And it'll be an entirely separate AI [11:21] ecosystem that Apple utilizes uh there [11:26] versus here because, you know, Google [11:29] can't operate in China. [11:30] >> All right, let's go. 250. [11:31] >> All right, [11:32] >> Michael. It's the 250th anniversary. [11:34] It's your piece. Our episode [11:38] 2. [11:40] >> This is our 2. Hold on. Yeah, we lined [11:42] it up. Hold that up. [applause] Good [11:45] job, Miss Nicole. So, 250th edition of [11:48] the Compounded Friends. [applause] [12:01] >> [music] [12:02] >> This episode is brought to you by [12:03] Federated Airmes. Active ETFs are [12:05] changing [music] the way portfolios are [12:06] built, giving advisers more flexibility [12:08] for their clients. But not all ETFs are [12:10] built the same. Federated Airmes puts [12:12] the [music] investments in their active [12:13] ETFs through a ruthless vetting process, [12:16] gaming out a wide range of market [12:17] scenarios, [music] [12:18] so only the strongest survive. The [12:20] result, a suite of 12 active ETFs [12:22] spanning [music] the full stock and bond [12:24] market. Whether you use them as core [12:25] building blocks or tactical allocations, [12:28] you'll get the strategies you want in a [12:29] convenient [music] ETF rapper. Simply [12:31] put, Federated Hermes has the active [12:34] ETFs to help you build [music] [12:35] portfolios designed to last because [12:37] they've been vetted for it. Explore the [12:39] full lineup at federated.com/us. [12:43] [music] [12:44] ETFs are subject to risk and may lose [12:47] value. Federated Securities Corp [12:49] distributor. Before investing, carefully [12:52] consider the fund's investment [12:54] objectives, [music] [12:54] risks, charges, and expenses. Read this [12:58] and more information in the prospectus [13:00] [music] or summary perspectus available [13:03] at federatedhermes.com/ [13:06] us. We are still celebrating the 250th [13:10] birthday, United States of America. [13:13] And we are going to cover a very special [13:15] piece of research that has come out over [13:18] the last week from our special guest, [13:20] returning champion, fan favorite, [13:23] [applause] [13:23] the legendary Michael Seas. Ladies and [13:25] gentlemen, [13:28] a little more. Crowd is going wild. This [13:30] man walked through the rain from 47th [13:33] Street to be with here with us today. [13:35] All right. Michael is the chairman four [13:38] block. Yeah. Michael is the chairman of [13:40] market and investment strategy for [13:42] [music] JP Morgan asset management where [13:43] he leads strategic market and investment [13:46] insights [music] across the firm's [13:47] institutional funds and private banking [13:50] business. He is the author of Eye on the [13:53] Market. I never miss it, a widely [13:56] [music] read commentary covering [13:58] markets, economics, energy, and policy. [14:02] Previously, he served for eight years as [14:05] chief investment officer JP Morgan's [14:07] global private bank after joining the [14:09] firm in 1987. Michael, welcome back. So [14:12] happy to have you here. How are you [14:14] feeling today? [14:14] >> I'm okay. [14:15] >> Uh are you a big patriot, medium [14:17] patriot? You're where would you put [14:20] yourself on the scale? Like are you are [14:22] you like Mr. America or like where do [14:25] you where do you stand? It's not a trap. [14:27] I'm genuinely curious. [14:28] >> It sounds like a trap. [14:28] >> No, no, no. [14:29] >> No. I [clears throat] I thought you were [14:30] you were a Patriot fan and that answer [14:32] to that was going to be absolutely not. [14:34] >> No, no, no, no. I know you're not. [14:35] >> But um yes, I [clears throat] am a [14:37] Patriot. Okay. [14:38] >> I am a Patriot. I'm I'm Yeah, [14:40] >> I asked that knowing the answer. The [14:42] tone of what you write. You're very o [14:45] over years. [14:46] >> It's very much um it's very much [14:50] debunking people that talk about Cell [14:52] America. It's very much about the [14:54] innovation in our economy, the dynamism, [14:56] etc. And obviously I know that's a a JP [15:00] Morgan's brand new building. They've got [15:01] an American flag blowing in the breeze [15:03] in the lobby. Um I've heard Diamond [15:06] Stump speech in person many times on TV [15:10] about why this is the greatest country [15:11] in the world. So I just wanted to kind [15:12] of get you on the record uh as a patriot [15:15] and someone who loves this country. [15:17] >> But I'm you know I'm analytically I'm a [15:20] patriot. Yes. This is a country with the [15:22] United States has the highest ratio of [15:24] people immigrating here versus the [15:26] people that leave still than any other [15:29] country in the world. Um and and the [15:31] data that I update every year with World [15:33] Bank data. So you know the people are [15:35] voting with their with their feet. [15:37] >> So would you say you're like an 8.7 or [15:38] >> Yeah. So like are you a 6.9? [15:41] >> Yeah. Look like a lot of people my [15:43] relatives came here built a live for [15:44] themselves and uh so [15:46] >> it still works. [15:47] >> All right. So, let's put this [15:49] >> warts and all. And look, and and in that [15:51] in that piece. Um, and you know, I I I [15:55] included a couple sections at the end [15:56] that my biggest long-term concerns in [15:58] addition to the to the unsustainable [16:00] level of federal debt are issues related [16:03] to the rule of law and um and and what's [16:06] going on with federal funding of uh [16:09] medical research. [16:10] >> Let's put uh let's put this image up to [16:12] set the set the scene. You titled your [16:15] special edition of Eye on the Market [16:18] Semiquentacles. [16:20] >> You got it. [16:22] >> So, it's the semi-quincentennial, [16:24] >> right? [16:24] >> But the point that you're making here is [16:26] despite all of the doubts and concerns [16:30] over the years about, [16:32] >> you know, when is it not going to be [16:34] America's moment anymore, right? [16:36] >> Or who's going to replace them or what's [16:37] going to replace the dollar. Despite all [16:38] of that, our tentacles are still very [16:41] much wrapped around the globe. [16:43] >> Yeah. I mean, I I I our clients have [16:47] benefited in in innumerable ways from an [16:51] asset allocation approach that more or [16:53] less since 2008 and March 2009 has been [16:57] overweight. The the dollar uh US credit, [17:00] US equities, US real estate, US [17:04] infrastructure, just across the board. [17:05] And you know there have been blips in [17:07] between but um uh I think there's people [17:11] are always seem to be looking for [17:13] opportunities to say take profits and [17:16] you know so far that hasn't been the [17:17] thing to do. Um when when you run the [17:20] numbers being long the US has been [17:23] enormously profitable. The only thing [17:26] that I've ever seen that matches the [17:29] profitability of being long the US over [17:31] the last 15 to 18 years was being [17:33] underweight Japan when it crumbled from [17:36] being 65% of global equities in 1989. [17:40] That's the only thing that compares. [17:41] >> So, uh, what makes it hard to be long [17:45] the US, long the dollar, long real [17:48] estate, long US stocks? What are what [17:51] why do we have so many people who are [17:53] like knee-jerk up that's it that's the [17:55] top for America or [17:57] >> well look because because on a cocktail [18:00] napkin the the PE ratios look very high [18:03] and they are very high but there's a [18:06] bunch of other things that you have to [18:07] look at which is okay but what's [18:09] earnings growth doing if you actually [18:11] look at the sectors that have the [18:13] highest valuations they have the highest [18:14] earnings growth so I a lot of your [18:17] listeners are familiar with a concept [18:18] called a PEG ratio IO which is just a PE [18:20] that they all love to look at and [18:22] earnings growth which they love to look [18:24] at separately. Why not merge them? And [18:25] if you look at the PEG ratios for the [18:27] US, they're not that out of line [18:29] compared to the rest of the world. Even [18:30] in the tech space, [18:31] >> who said the who said the quote, I don't [18:33] I don't need to pay an analyst to tell [18:35] me that a stock at eight times earnings [18:37] is cheap. I need somebody to tell me [18:39] when a 40 times earnings stock is cheap. [18:41] >> I love that. No, it's expensive. No, the [18:44] opposite. [18:44] >> The opposite. When eight times is [18:45] expensive and 40 times is cheap. [18:47] >> I don't know who said that quote. It's [18:48] but it's it's a wonderful one. [18:49] >> It's a good one. Right, [18:50] >> Michael. My favorite part about [18:51] >> I'm sorry. And the dollar, too. I mean, [18:52] we'll get into that, but like the dollar [18:54] is everybody's punching. [18:55] >> It's the dollar. It's the dollar. The [18:56] deficit, the unsustainable nature of of [18:58] all of this and fiscal policy and [19:00] monetary policy. Whenever people talk [19:02] about debt to GDP and all the problems [19:04] that they have with with what's going [19:06] on, they never mention an alternative. [19:09] Okay, fine. So, let's just humor them, [19:10] right? Let's just say that the US is on [19:12] the ropes and it can't maintain its [19:14] reserve currency status forever. what's [19:16] going to replace it? And you did a very [19:18] good job not only lining out why the [19:21] dollar is still the dominant currency, [19:23] but all the reasons why maybe the [19:25] Chinese alternative is completely [19:27] absurd. [19:28] >> Yeah. Well, look, I a day will come in [19:32] our in in our lifetimes [19:34] when, you know, there the piper has to [19:36] be paid on some of these things and we [19:38] can talk about what that might look [19:39] like. But my job as an analyst in in [19:42] managing the firm's money and we oversee [19:45] $4 trillion on behalf of institutional [19:47] individual investors is to make sure [19:49] that we don't run for the hills too soon [19:51] and leave enormous profits on the table. [19:53] That's what's driving this. So the first [19:55] discipline you have to have is everybody [19:58] thinks that the dollar is is due for a [20:01] collapse is are there any signs it's [20:03] happening? Are there any signs it's [20:04] happening right now? And so we have all [20:06] of these metrics we look at in terms of [20:08] the dollar share of world trade, the [20:09] dollar share of foreign exchange [20:11] allocations, the dollar share of swift [20:13] payments, the dollar share of [20:14] international currenc. [20:19] So um if it's the table on the top [20:22] there, the dollar share of crossber [20:23] loans, [20:24] >> what you're showing is yearbyear there's [20:26] really no change but not just stable, [20:28] dominant. [20:29] >> May so yes, maybe one day. Yes, dominant [20:31] because the US is only 20% of GDP and [20:34] most of these numbers are much higher [20:35] than that. You know, roughly double. So [20:38] maybe one that day that will change. But [20:40] I'm always struck by the lack of the [20:42] discipline for the dollar bashers to [20:44] even acknowledge that none of the [20:47] pillars of the dollar as the world [20:49] reserve currency are shifting. And it's [20:51] kind of easy to come out and give like a [20:53] rubinius-esque, you know, trashing of [20:55] the dollar, but you should have some [20:57] discipline to say, are there any signs [20:59] that it's happening? Who's gaining share [21:01] on the bottom left? Nobody. I mean the [21:03] other side. [21:03] >> It's not the yen, the pound or the or [21:05] the yuan. [21:05] >> And the ones on the right are currencies [21:07] that are I mean these are very small. [21:09] >> They are they're microcurrens. [21:10] >> None of these none of those could be [21:12] reserved. So So Japanese yen, great [21:15] British pound, Chinese yuan. There's a [21:16] lot of stability in this chart at least [21:19] over the last 5 years. Nobody's really [21:21] gaining anything, right? Is that okay? [21:22] >> Yeah. And and you've seen small [21:24] allocations into the Sing dollar, [21:26] Norwegian croner, stuff like that. But [21:28] the bottom line is that the the things [21:31] that make the dollar the world's reserve [21:33] currency aren't really changing. And [21:35] that's that's the important takeaway. [21:37] And like maybe one day that will change, [21:39] but so far there's no sign of it. So the [21:41] last thing we're going to do at this [21:42] point is make major structural changes [21:45] to the currency overlay in the [21:46] portfolios that we manage. [21:47] >> Yeah. So one of the things that the [21:48] dollar bears harp on is the buying power [21:51] of the dollar. They act as though things [21:54] don't cost more in other currencies. [21:56] also is 0.1 and then point 2 they [22:00] they'll say oh the dollar has lost 96% [22:02] of its purchasing power over the last 50 [22:04] years or whatever they say um which it's [22:07] a tortology yeah the price of everything [22:09] has gone up and okay but then the [22:11] problem is so you say to them okay well [22:14] those are the same people though that [22:16] also would tell you to stay out of the [22:17] stock market you can't be a dollar bear [22:19] and a bear on stocks because stocks have [22:22] more than helped you keep your [22:24] purchasing power and it's the only [22:25] answer. [22:25] >> Yeah, your dollar does not belong under [22:27] your mattress. [22:28] >> Real estate and stocks are the only [22:29] answer if you're if you're worried about [22:30] the person power of the dollar. [22:32] >> Now, if you're worried about US [22:33] fundamentals over the last few years, [22:34] the places that people have gone is gold [22:36] and crypto. And so you know what one of [22:39] the things that we watch is that I think [22:41] is a key arbiter is when is the GIC when [22:44] are the big institutional investors to [22:47] MASC you know and the big sovereign [22:49] wealth funds going to make strategic not [22:53] just tactical but strategic allocations [22:55] to digital currencies in their foreign [22:56] exchange reserves. That would be a [22:58] seismic shift and so far for the most [23:00] part that has not happened. I also was [23:03] fascinated to see Worsh's initial public [23:08] statements because he listen I mean [23:10] think about this process. Trump's been [23:12] out saying that he wants the funds rate [23:14] below 1% by the end of this year. Um and [23:17] and other similar things and you can [23:19] kind of see a confrontation brewing that [23:21] that rhymes a little bit with the [23:22] confrontation that Nixon had with Arthur [23:24] Burns. Um and but Wars has kind of come [23:28] out with with some statements that [23:30] reinforce historically that he's a hard [23:33] money guy that's not going to get pushed [23:34] around. [23:35] >> Khi is at 53% chance of a rate hike this [23:38] year, which is un I think roughly [23:40] unchanged from before WS actually got in [23:43] the seat. And so [23:45] >> and we started the year with the markets [23:46] pricing in and cuts. [23:48] >> Yep. So, you know, so far he and and I I [23:52] don't think it's a coincidence that some [23:53] of those comments have coincided with [23:55] some of the steam coming out of of [23:57] crypto and gold. [23:58] >> One of my favorite charts, speaking of [23:59] gold, you see this all over the place, [24:01] >> the amount of money that central banks [24:04] have in gold. [24:06] >> Let's show you chart for Daniel [24:08] >> is all right. Well, maybe this is a [24:11] price issue. So, so let's let's say you [24:14] have a guy and he's he's got 100,000 in [24:17] cash and and a house worth $100,000 and [24:20] all of a sudden Mark Andre comes along [24:22] and wants to buy his house for a million [24:24] dollars. His asset allocation now looked [24:27] like it's 90% real estate. But that's [24:30] simply because of a price change, not [24:31] because he bought more real estate. That [24:33] is a very simple metaphor for what's [24:35] happened to the gold market. Gold prices [24:37] have gone up. So the value of gold and [24:39] foreign exchanges has gone up. the the [24:41] actual allocated troy ounces of gold [24:45] that these the central banks own has [24:47] fallen [24:49] >> is in valuation terms and that's the [24:51] gold line in this chart. So gold has [24:53] gone from like 1,500 to to 4,000 [24:55] something an ounce. [24:56] >> And that's that is 95% of the reason why [25:01] gold as a share of foreign exchange [25:02] reserves has gone up as much. It's not [25:04] because the central banks are kind of [25:06] chasing it and hoarding. [25:07] >> But they write these articles as though [25:09] the central banks are loading up on more [25:11] gold. No, all that's happening is the [25:13] gold that they currently own. [25:14] >> I mean, is it any surprise that the kind [25:16] of gold bugs are misrepresenting the [25:19] data? It goes with the territory. [25:21] >> Can't believe it. So, um, this this was [25:24] amazing. You wrote that. All right. So, [25:26] China's maybe another potential, right? [25:28] The Chinese economy is huge. If it's not [25:29] us, perhaps it's them. You said if China [25:31] fully opened its capital account, [25:33] possible outflows could crush the R&B [25:35] and trigger a collapse in Chinese equity [25:37] real estate markets. Even with the [25:38] controls, China reportedly experienced a [25:42] record 800 billion to a trillion in [25:45] capital outflows in 2025. [25:48] >> Yeah. I mean, China does a lot of [25:51] amazing things, right? There's a lot of [25:52] innovation in China. They're catching up [25:54] to the US in terms of in terms of [25:56] patents and all sorts of experiments. [25:58] And so, you know, when you say bad [26:00] things about when you when you say [26:01] critical things about certain aspects of [26:02] the Chinese economy, you know, there's a [26:04] lot of like every country, there's good [26:05] things and bad things. One of the [26:07] undeniable aspects of the Chinese [26:08] economy is in addition to personal [26:10] surveillance, they have monetary [26:12] surveillance. You can't get money out of [26:13] the country. And the M2 to GDP ratio in [26:17] in China is off the charts compared to [26:20] other countries. And that was one of the [26:21] key charts. [26:22] >> M2 the money the money supply, [26:25] >> right? [26:25] >> So So you're looking at [26:26] >> because it can't leave. [26:28] >> Yes. Not this one. [26:29] >> Not this one. Okay. So maybe we [26:30] >> But let's let's let's look at this one. [26:32] Let's play. For many years, my wife was [26:34] on the board of Sesame Street. Um yeah, [26:36] she for 20 years. She loved the show and [26:39] um she kind of Yeah. Well, she used it [26:42] to kind of teach me spelling. And so, [26:45] um, let's in Sesame Street, they used to [26:47] have, you know, one of these things is [26:48] not like the other, right? So, that's [26:50] what this chart is. So, you don't have [26:51] to be a mon, you know, macroeconomist to [26:53] understand this. This looks at the money [26:55] supply, central bank assets, and foreign [26:58] exchange reserves for three different [26:59] kinds of places. The first one are [27:02] pegged currencies, um, you know, like [27:05] the Hong Kong dollar and the Singapore [27:07] dollar, things like that. [27:08] >> They peg they peg the value to the [27:10] dollar. and they manage the money supply [27:12] to keep that in line. And then the [27:13] middle one is and by the way the first [27:15] one also includes some of the Gulf [27:17] currencies that are pegged. The middle [27:19] one is the Mexican peso and the one on [27:21] the right is China. [27:23] >> So one of these things is not like the [27:24] other. And so in China, the growth in [27:27] the money supply has massively [27:29] outstripped the accumulation of central [27:30] bank assets and foreign exchange [27:31] reserves, which tells you if they ever [27:34] loosened the foreign exchange controls, [27:36] >> that's all. That's all excess money that [27:39] would immediately leave. [27:40] >> Immediately leave, which could crush the [27:42] and they would never do it. [27:43] >> Goes into what it's already secretly [27:44] going into now, Vancouver real estate, [27:47] like like things like that or [27:48] >> all of the all of the above, right? I [27:50] mean, and and there's no I mean, I don't [27:52] think it's really clear where exactly it [27:54] would go. We just know that it would go [27:55] someplace. [27:56] >> Okay. [27:57] >> And so, and and the the reason I put [27:59] this in there is any you know, the [28:02] definition of of the world's reserve [28:03] currency is that you don't have that [28:05] kind of sort of damicles hanging over [28:07] your head. And so, China is not a [28:09] candidate to be the world's reserve [28:11] currency if they have this kind of [28:12] structure. [28:12] >> It would go into dollar. How do they not [28:14] have with that explosion in the money [28:16] supply and all of that money captive in [28:18] the Chinese economy? How do they don't [28:21] how do they not have significantly worse [28:23] inflation? [28:24] Aren't like don't we think [28:26] >> the real estate collapse people are [28:28] >> so they have def enough deflation in [28:29] real estate that it's it's stopping that [28:33] money supply growth from from becoming [28:34] inflation [28:35] >> and and remember Chinese there's massive [28:37] amounts of Chinese saving. [28:39] >> Yeah. [28:39] >> In fixed income instruments and things [28:41] like that. [28:42] >> Okay. Yeah. like insurance [28:44] >> the China ch China is like the United [28:46] States 100 years ago there's little to [28:49] no uh safety net in terms of Medicaid [28:51] Medicare and things like that so people [28:53] have to precautionarily save the high [28:55] savings rate [28:55] >> and they have very and they're having [28:56] increasing life expecties and no safety [28:59] net so people have to precautionarily [29:00] save to do that [29:01] >> okay here's the other one they're [29:03] dumping our bonds slash they're going to [29:06] start they're going to stop buying our [29:07] bonds uh the bond the bond global bond [29:10] vigilantes chi China. We went from China [29:15] owns us was the thing that like late [29:17] night comedians would say on state like [29:20] we ow like uh we're we're debtors to [29:22] China. That was the problem. Now it's [29:25] the opposite. China's selling all our [29:26] bonds. They're dumping our debt. What do [29:28] they know that we don't? [29:29] >> So let's let's let's tackle this this [29:31] concept. So [29:33] again, like the premise that global [29:36] investors would respond negatively [29:40] to this administration and its policies [29:42] is certainly plausible. There's just not [29:44] a lot of evidence that that's actually [29:45] happening. Right. So I don't start out [29:47] with a kind of a value judgment or a [29:49] bias against some of these statements. [29:51] All I want to do is see whether or not [29:52] there's any reality to them. [29:54] >> Yeah. And from what we can tell from the [29:57] chart on the upper right and on the [29:58] bottom, foreigners are still [30:01] accumulating treasuries and T bills, [30:04] corporate bonds, and equities at at [30:07] roughly the same pace they have been. [30:09] >> But the share is going down. Why? [30:10] Because we're buying more of them now. [30:11] >> No, because we're issuing that we're [30:12] issuing debt so much more quickly. [30:14] >> So that's so that's worrisome potential. [30:16] >> It is. It is. There's a lot of debt. So [30:18] the share has gone down. But Russia so [30:20] far is really the only country in the [30:22] world that said, you know, we're going [30:24] on a buyer strike and we're going to [30:25] sell our treasuries and agencies. [30:27] There's no evidence that that's [30:28] happened. People always get [30:30] hoodwinkedked. The the Chinese numbers [30:32] that are officially reported have gone [30:34] down, but a guy named Bradzer at the [30:36] Council on Foreign Relations has done [30:38] all the work on this. The they're buying [30:40] a ton of them through this Belgian euro [30:42] clear account. And when you adjust for [30:44] the two of them together, they're [30:46] they're treasury. have to buy them [30:48] because they need [30:50] >> they don't have much of a choice. [30:51] >> They need first of all there's you run [30:53] out of places for the money to go. [30:55] >> Yeah. [30:56] >> But second of all they conduct so much [30:58] trade in dollars they need to have [31:00] access to it. [31:00] >> Well essentially they run a large [31:02] capital account surplus and so that that [31:04] capital has to get invested someplace [31:06] and [31:07] >> even if politically they wanted to do [31:09] that [31:11] they can't there's a limit to what they [31:12] can do. [31:13] >> That's right. [31:13] >> Okay. Do we should we care whether or [31:16] not China is holding our bonds, buying [31:18] our bonds? It's just a thing that people [31:20] >> I think so. [31:21] >> We we should like investors should care. [31:23] >> Yeah. Because it's unclear. Um I I I [31:27] that I would rather have that than [31:30] China decides to bail out of $500 [31:33] billion and to prevent yields from going [31:36] up best and has to strongarm the banks [31:38] to buy them, which is how the European [31:40] banks function. like I'd rather not have [31:42] the US banking system acting as a [31:45] backdoor placement agent for US traders. [31:48] Is there a danger of that with with if [31:50] the if the the tariff rhetoric gets [31:52] worse or like is it is there a [31:54] legitimate danger to that do you think? [31:56] >> China is buying as you mentioned China [31:58] is buying all those treasuries and [31:59] agencies in its own self-interest [32:01] nothing to do with us and so and if that [32:03] self-interest ever changed they would [32:05] allow them to advertise down and or sell [32:07] them and then you know we would have to [32:09] see how much other demand there is for [32:12] that long duration. Have you looked have [32:13] you looked at like the uh sovereign [32:15] wealth funds from the oil producing [32:18] countries uh and whe whether or not [32:20] they're potentially taking the place of [32:22] the Chinese buyer? [32:24] >> They're they're the pace of buying is [32:25] roughly the same there. There no major [32:27] changes there. [32:28] >> Okay. All right. [32:29] >> So we still got it. [32:31] >> I mean we are we are as as the as the [32:34] world's reserve currency. We still have [32:36] what is referred to as this extraordin [32:38] exorbitant privilege of being able to [32:41] run large deficits and have people buy [32:43] the bonds without it without worrying [32:45] about what happens to yields. [32:46] >> So do you think there would be a a [32:48] reckoning that would happen as a result [32:51] of a individual event? Would it be a [32:54] slow degradation of the transactions? [32:57] >> I think that um and I mentioned this in [32:59] the piece. My concern is in 2031 or 2032 [33:03] you reach this crossover point where [33:07] 100% of federal tax revenue has to use [33:09] to get pay to pay interest on the debt [33:11] and entitlements with nothing left or [33:12] anything else. That is that is a scary [33:15] day. And two two or three years before [33:18] then I would expect the rating agencies [33:21] to threaten to downgrade the United [33:23] States unless there's some serious [33:24] restructuring of the entitlements. [33:25] >> I mean that's a ticking time bomb in [33:27] real life. [33:27] >> That's not far away. [33:29] >> Yeah. I think it's I think it's three [33:30] years away. [33:31] >> So, three years from now, the ratings [33:33] agencies are going to look to two years [33:36] later. [33:36] >> That's right. They have they have to [33:38] give the US a chance to say, "Look, [33:39] you've got a couple of legislative [33:40] sessions coming up." Um, and you know, [33:43] uh, and we all know what that blueprint [33:45] looks like. B Simpson published it, you [33:48] know, 20 years ago. We know there's only [33:50] there's only four or five big policies [33:52] that can move the needle, and we all [33:53] know what they are. [33:54] >> Tax tax hikes. Well, the fairy tale is [33:57] we just grow our way out of it. But is [33:58] that [33:59] >> and that did happen under Eisenhower. [34:01] >> Yeah. [34:02] >> And um under Eisenhower, you know that [34:05] he inherited a debt to GDP ratio that [34:07] wasn't that different than what we have [34:08] now. It's about 100% of GDP and it fell [34:10] to 60 by the end of his 8-year term. [34:13] They didn't raise taxes. They didn't cut [34:15] spending. They grew like crazy because [34:17] he had progrowth policies. Um yeah [34:21] that's what the [34:22] >> the difference is he inherited that as a [34:24] consequence of we defeated the empire of [34:27] Japan and the Nazis [34:29] >> right no but the point was that the [34:30] decline like you said the decline in [34:32] that debt to GDP ratio happened because [34:35] nominal and real growth was high [34:37] >> right [34:38] >> so if you if you if you want to replay [34:40] that playbook [34:42] um that's kind of what Trump was running [34:44] on before he went into this nivist [34:48] territory [34:49] um And why the business, you know, the [34:51] business roundt and a lot of business [34:53] people and people in Silicon Valley were [34:56] initially [34:58] enamored with the policy mix they [35:00] thought they were getting because they [35:02] thought they were getting a version of [35:03] that. Let's let's progrowth, you know, [35:06] let's build, you know, everything and [35:09] and um and it just kind of shifted from [35:11] there. Um, but I also don't have a lot [35:14] of confidence that a Democratic [35:16] administration would prioritize growth [35:18] and I think a lot of other things would [35:19] get prioritized instead. [35:21] >> What if we take a 5% stake in Open AI [35:23] like like some people are suggesting, [35:25] sell it to CO2 at a billion a trillion [35:27] dollar valuation? Boom. 50 billion [35:29] bucks. Dump it. [35:30] >> Something tells me that would be a a [35:34] negative market indicator rather than a [35:35] positive. [35:36] >> Are you surprised that given all the [35:37] deregulation which they promised and [35:40] they actually delivered on? Yeah. [35:41] >> Uh I would say [35:42] >> they have they have been [35:44] >> for sure. [35:44] >> Yeah. [35:45] >> Are you surprised that with all of that [35:47] >> and the Fed having eased in 22 and 23 [35:52] and just uh generally getting an [35:55] extension on the 2017 TCJA last year [35:59] like all of the things that they've [36:01] actually done, we're still talking about [36:03] 2.8% GDP growth. [36:05] >> Yeah. [36:05] >> Are you surprised? [36:06] >> Well, it it um deregulation is a [36:09] slowmoving thing. it takes a while. Uh [36:12] so when you look at deregulation of the [36:14] airlines and um telecom in the 1990s um [36:19] Clinton was the beneficiary of that but [36:21] it took three to five years for that [36:23] really to kind of mobilize the private [36:25] sector to do things differently than [36:27] they had been and those things take [36:29] time. Okay. [36:30] >> Um and then you know at the same time [36:32] that the administration is pushing this [36:35] deregulation they're also shrinking the [36:37] labor supply. Uh they're also increasing [36:39] the cost of doing business through [36:41] tariffs. Um one of the interesting [36:43] things about what [36:44] >> they're offsetting it with the other [36:45] things. [36:46] >> Oh yeah, there's a lot of offsetting [36:47] policies there and it's not always clear [36:48] to me that they understand those [36:50] offsetting forces. [36:51] >> Um you know the the around 80% [36:56] of the kind of AI trade are are subject [37:00] to tariff carves out carveouts, right? [37:02] So by hook or by crook around 80% of [37:05] semiconductors and related things are [37:07] not subject to these tariffs but only [37:09] 20% of the kind of power buildout is [37:12] subject to those exemptions. So the and [37:16] that's where the bottlenecks are. So the [37:18] cost of transformers is going up. I mean [37:20] you know you know like the deficit clock [37:22] that keeps going. I have a version of a [37:24] deficit clock in my head except it's the [37:27] price per megawatt of a combined cycle [37:29] gas turbine which doubled over the last [37:32] three years and is still rising [37:33] >> and you can't have one anyway. [37:35] >> It'll take you [37:36] >> and you need three years before you're [37:37] on your top of the list. [37:38] >> I mean, [37:39] >> yeah, people have actually talked about [37:41] um invoking the Defense Production Act [37:43] to kind of compel uh G Vernova to [37:46] produce more of them. They've announced [37:47] a partial expansion but not kind of a [37:49] wholesale increase in production. I [37:51] don't blame them. Um, it's really just [37:53] them Mitsubishi and Seammens that make [37:55] these things, which is why there's so [37:56] much demand for repurposed uh shipping [38:00] engines for Mortzilla and and airplane [38:02] engines that people are stringing [38:04] together in daisy chains to to try to [38:06] get this and why people all of a sudden [38:07] have bid up uh Bloom Energy because they [38:10] make these solid oxide fuel cells that [38:13] can convert natural gas into power. And [38:15] so, you know, but that that's the [38:17] administration. some of the policies [38:18] have hurt and gone against the uh [38:21] >> I guess that's the answer for how we [38:23] have these American companies reporting [38:24] earnings that are up 70 80 100% [38:28] year-over-year [38:29] and then you just look at like overall [38:31] economic growth and you just don't see [38:33] the same sort of momentum [38:35] >> right well remember around 70% of GDP is [38:39] consumption right this is we we've all [38:41] this we all started studying this as [38:43] markets people in the early9s [38:44] >> and that's not going to grow [38:45] significantly faster no matter 70% of [38:48] GDP is consumption, but roughly 60 to [38:51] 70% of S&P profits is production. So the [38:54] S&P is a production index and the [38:56] economy is a consumption index. Which is [38:58] why when people say, "Oh my god, look [39:00] how well the stock market's doing. It [39:02] doesn't make sense relative to the [39:03] economy." Those kind of gaps can persist [39:05] for quite some time. And the reverse can [39:07] happen too. Look what happened in the in [39:10] 2001. The economy was was doing fine. [39:13] the the the NBER wasn't even sure until [39:16] years later that there was a recession [39:17] at all and we had a 40% decline in the [39:20] stock market. So these these things can [39:22] disconnect. [39:22] >> Michael, you have access to the smartest [39:25] analysts in the world and I'm sure your [39:26] clients are asking this all the time. [39:28] What's going on with the memory stocks? [39:30] So Daniel Chart 16, please. This is from [39:32] Joey Politano. We are now importing more [39:35] uh from Taiwan than we are from China, [39:37] which was that would have been [39:39] completely unthinkable just a couple [39:40] years ago. terms. Yeah. [39:41] >> Just completely unthinkable. [39:44] >> Um and the question is as semiconductors [39:47] are now 18% of the S&P 500, whatever it [39:49] is. [39:49] >> Yeah. [39:50] >> Are these stocks these companies that [39:53] were historically absolutely cyclical, [39:56] >> have we removed the cyclicality? Because [39:58] to me, this this is like the whole thing [39:59] over the next couple years where the [40:01] stock market goes. [40:02] >> I [clears throat] I I thought I was fine [40:04] with this [40:06] until about three months ago. You'll [40:09] find with what the semis being as big as [40:11] they became. [40:12] >> Yeah. [40:12] >> Proportionally to other stocks. Yes. [40:14] >> Like Micron's now bigger than Meta are [40:16] right there. [40:17] >> Yeah. I mean to me [40:20] my my concern is not [40:25] not the performance of these stocks but [40:27] the perform but the performance relative [40:29] to what what's going on around it. So [40:31] let me explain what I mean. in the in [40:34] [cough and clears throat] the in the [40:35] from 95 to 99 the comm equipment stocks [40:40] were tracking the comm services stocks. [40:43] And so the the the market caps and the [40:46] earnings growing up together. And you [40:48] started getting this signal in early 99 [40:51] where Cisco and the other infrastructure [40:53] stocks were still zooming [40:55] but the you know the ISP providers and [40:58] those kind of companies started to roll [41:00] over and the smart people I remember [41:03] Gary Brinson I don't know if you [41:04] remember him but some of the smart [41:06] market analysts at the time that saw [41:08] this coming um said wait a minute you [41:11] know who's paying for all the [41:12] infrastructure companies whose market [41:15] caps and earnings are rolling over [41:16] >> the stock prices of the buyers were [41:19] plunging. There's no way they're going [41:21] to buy as much of this equipment next [41:22] year because the stock market investors [41:25] are telling them not to by selling their [41:27] stock. and and you saw and if you you [41:29] know when you if you wanted to look even [41:31] more closely you would compare Cisco and [41:33] Verizon and like Verizon stopped going [41:36] up and Cisco kept zooming and like it [41:39] was clear who is Cisco going to sell [41:41] this stuff to and the last three months [41:45] >> the hyperscalers are starting to roll [41:47] over even though the semiconductor index [41:49] is zooming at the end of the day you [41:51] know that's if if this if the if the big [41:55] four hyperscalers plus oracle [41:58] can't convince the market that they're [42:00] earning a good return on a trillion five [42:02] of capital spending outlays. Those [42:05] infrastructure stocks are at risk at [42:06] some point. So, I'm starting to feel a [42:08] little nervous about where we are right [42:10] now. [42:11] >> I remember I remember the optical stocks [42:14] were the last to get the memo. [42:16] >> Yep. [42:16] >> Cuz they're [42:17] >> Oh, their order book looked great, [42:18] >> right? They're in the back of the train. [42:20] They don't know that the locomotive in [42:22] the front of the train just hit [42:24] something or went off a cliff. they will [42:26] be the last car off the cliff. They [42:27] won't see it coming. All they can see is [42:29] the car in front of them, right? [42:31] >> So, if you're a component supplier to [42:33] this, I don't give a [ __ ] what your [42:35] guidance is, [42:36] >> right? [42:36] >> Cuz you're not [42:37] >> And chances are [42:38] >> you're not going to be the first to [42:39] know. [42:40] >> Chances are before this is done, they'll [42:42] they'll all announce a production exp [42:44] capacity expansion, [42:45] >> right? [42:45] >> Did Micron do that today that they're [42:47] investing in America to produce more? [42:50] Well, that's kind of that's a [42:52] >> Micron just signed a signed a deal a [42:56] massive upstate New York facility. [42:58] >> That that's that's kind of rhymes with [43:01] the you know the TMC project [43:04] semiconductor projects in Arizona. It's [43:06] it done for geopolitical reasons. [43:09] >> Okay. Um, [43:11] >> I I I give the administration credit for [43:14] strongarming [43:16] foreign countries to finally commit FDI [43:20] in the United States. Um, that's what [43:22] other countries have been doing forever. [43:24] And so I think they're they're I I like [43:25] the way they're doing that. [43:26] >> Do you think Apple's lobbying efforts at [43:29] getting the White House to approve them [43:32] buying memory from Chinese suppliers uh [43:35] has a has a chance of working? And even [43:37] if it does, could it potentially be [43:39] meaningful or are we just going to be [43:40] looking at this memory supply constraint [43:43] for as far as we can see? [43:47] >> You know, these I [clears throat] mean [43:50] sub 2 nanometer semiconductors and the [43:53] really good high bandwidth memory stuff, [43:55] the the leading companies have so you [43:58] know have such an economies of scale. I [44:00] I think it's it's it's hard to imagine a [44:03] world where they [44:04] >> it's not going to be this sudden bounty [44:06] of supply, [44:07] >> right? [44:08] >> Michael, I'm curious about where this [44:09] warning label came from. So when you [44:11] started to write about the AI stuff, you [44:15] wrote this section includes technical AI [44:17] jargon, conjecture on AI products and [44:20] services that are rapidly changing views [44:22] on non-public companies whose [44:23] disclosures I mean you go on you say if [44:25] you have messianic opinions on these [44:26] topics and cannot stand to see other [44:28] points of view or if you are shocked the [44:30] topic X or Y did not get mentioned. Um [44:33] and whatever who are you talking to? Um, [44:36] I get, you know, a lot [44:38] >> people who email you. [44:39] >> I get I get fan mail. [44:40] >> This is a preemptive shut the up. [44:42] >> Well, no, I I get a lot of fan mail and [44:44] on different topics and some of it's not [44:46] so fanny. Um, and um, you know, look, at [44:50] the end of the day, I expect our clients [44:52] to judge us based on our portfolio [44:54] performance, you know, not on the things [44:55] that I say on podcasts. But um I find [45:00] that the people that exist in this [45:03] universe have more passionate [45:06] unbridled [45:07] one-sided opinions [45:10] as as any sector that I cover and that [45:12] includes energy. [45:14] >> You're saying the AI the people that are [45:16] focused on AI, [45:17] >> right? Right. You know, debates about [45:18] token prices, Jivvon's paradox. I mean, [45:21] these people can really get themselves [45:24] into a tizzy. And I think that's because [45:26] a lot of them spend a good chunk of [45:28] their time on Twitter and other quasi [45:32] anonymous platforms where they can [45:34] scream and rant at people anonymously [45:36] and you know aren't used to actually [45:39] having discussions with people in the [45:42] room. [45:42] >> Um and so um people have really really [45:45] really strong opinions about this kind [45:47] of stuff. Um, you know, Michael [45:50] [clears throat] Barry is getting Barry [45:52] Barry is is is getting people very [45:55] excited and I think his research is [45:56] really interesting, right? [snorts] How [45:58] long are some of these products being [46:00] depreciated? [46:01] >> Um, and and and does that match up [46:04] against their actual useful lives? He's [46:06] asking a lot of good and interesting [46:07] questions, but I think people people [46:10] sometimes are getting very very carried [46:12] away. [46:12] >> So, you called this the most technically [46:14] complex market catalyst in the last 40 [46:17] years? Without question. [46:18] >> So say more. [46:19] >> Well, I mean if we think [clears throat] [46:21] about the market catalysts of the last, [46:23] you know, in the 30 years that I've been [46:25] doing this, you know, B2B software, the [46:28] fiber buildout, I mean, [46:29] >> looks quaint. [46:30] >> Yeah. I mean, you you middle school [46:32] level education could equip you to [46:34] understand what was going on. Now all of [46:36] a sudden you have to kind of start with [46:37] Google's foundational transformer paper [46:40] in 2018 um and and go on from there in [46:43] terms of how these things are actually [46:45] functioning and you know what does it [46:47] take to produce some of these DRAM and [46:51] uh GPUs and how do they compare with [46:53] TPUs and CPUs their energy intensity. I [46:55] mean, these are some pretty complex [46:57] technical things, right? And so, [46:59] sometimes you'll have a company that [47:00] comes out and announces that they were [47:02] able to do something and, you know, it [47:04] takes quite a bit of work to unravel it [47:06] and see whether or not that story is [47:07] real. [47:08] >> The problem is the market's reacting [47:09] immediately regardless. Right. [47:11] >> Right. The the market is making up its [47:12] mind really quickly. [47:14] >> Right. Uh the market's also getting a [47:15] lot wrong because of the rapidity of the [47:19] of the changes that are happen like uh [47:21] market was very convinced that Google [47:24] was about to either cannibalize its ad [47:26] business or just in general have the [47:29] floor wiped uh with it because of Open [47:32] AI. And then the market did a total 180 [47:35] on that story and they said, "Oh, [47:37] actually, uh, Gemini is superior to a [47:40] lot of these other LLMs and Google's [47:44] user base is going to enable it to [47:45] monetize faster than anyone else." And [47:48] then you just have like this complete [47:50] 180 in that name. And that's one of like [47:52] many examples I could come up with. Do [47:54] you remember a couple years ago people [47:56] were panicked about the fact that um [48:01] Apple had this default agreement that [48:03] Google Google was going to be the [48:04] default browser and in Europe all of a [48:07] sudden they said no you know people have [48:09] to choose the default browser that and [48:12] and they ended up picking Google anyway [48:14] right I mean [laughter] [48:16] >> you know when's the last time somebody [48:18] said hold on I'm going to Bing that like [48:20] nobody does that [48:21] >> I'm going to use duck do go to search to [48:24] search for something, right? Of course. [48:25] Okay. Uh I just I feel like um the more [48:30] the the more rigid you you are about how [48:33] this is all going to develop, the less [48:35] able you'll be to make money, [48:36] >> right? [48:37] >> Because of how quickly everything's [48:38] moving. [48:39] >> I am concerned though that [48:42] for the last few years, [48:45] the big hyperscalers have been financing [48:47] their capital spending out of internally [48:49] generated cash flow. And just towards [48:51] the end of last year they started [48:54] borrowing to do it. Now so far the [48:55] borrowing numbers are small and you have [48:57] to have the discipline to look at the [48:58] debt to Ebbitar ratios and so far they [49:01] look okay but the trend the first [49:02] derivative of the trend is shifting away [49:04] from internally generated cash flow and [49:07] and some of the charts that we had in [49:08] the deck in the piece [49:10] showed that the free cash flow [49:12] projections are for the hyperscalers [49:15] >> are heading towards are heading very [49:17] low. But don't you think the market's [49:19] being sober about how they're treating [49:21] the equity of these companies? The [49:22] market is saying, "Yeah, we don't really [49:23] love this." In fact, [49:24] >> yeah. And that's the reason why the [49:25] hyperkills haven't been performing well, [49:27] have not been performing well. Look at [49:28] the chart. The chart at the upper left [49:30] is essentially the capital spending in [49:31] R&D as a share of revenues and the chart [49:34] at the upper right is the free cash flow [49:36] margins. You know, this is this is these [49:39] are some pretty steep declines. And so [49:41] within 18 months, the hyperscalers are [49:44] going to have to demonstrate that JP [49:47] Morgan and the and the big corporate [49:49] entities in the United States are going [49:52] to be rapidly increasing their AI [49:55] adoption and paying for it and [49:57] [clears throat] not scrambling to use [50:00] free Chinese free Chinese models. [50:03] >> Yeah. And [50:05] I can tell you within JP Morgan, there [50:08] are lots of CFO meetings about [50:10] escalating token costs and what should [50:12] be done about it. And so, um, [50:16] >> my understanding is that we're getting [50:17] more efficient with token use. The [50:20] problem is the aggregate demand is [50:23] ramping so fast that it doesn't matter [50:26] that the cost of a token is falling [50:29] because once you give people these [50:31] tools, the only thing they can think [50:33] about is what else can this do? [50:35] >> Yeah. [50:35] >> And now you have corporations imposing [50:37] hard caps on employee use of AI. [50:40] >> I [50:40] >> I'm more so I'm more worried about that [50:42] than about I don't think any of this [50:44] goes back in Pandora's box. I it's not [50:47] 3D printing where people play with it [50:50] and they say ah you know what we don't [50:51] really need that [50:52] >> right [50:52] >> like nobody is turning the other way [50:55] >> the agentic AI stuff is is extremely [50:57] powerful and um you know I think I think [51:00] that's going to continue to grow and [51:02] those are massive token consumers but [51:04] again like there's going to be a lot of [51:06] pressure [51:08] in the same way and same thing's [51:09] happening in software right um every CTO [51:13] is being asked [51:16] Why can't we use some of these tools to [51:18] displace some of the vendor software [51:20] relationships we have? And even if we [51:21] can't, can't we call this the our [51:24] vendors and threaten to do it unless [51:26] they cut the, you know, our our [51:28] subscription fees in half? [51:29] >> Yeah. [51:29] >> So, there's there's a lot of [51:31] >> This is the first time This is the first [51:33] time that these contract renewals aren't [51:36] let layups. [51:37] >> Yeah. [51:38] >> At at a at a built-in 3% rate increase [51:41] or whatever it is. [51:41] >> They're automatic escalators, as far as [51:43] the eyes can see. I wrote a a I write an [51:46] alternatives review about, you know, [51:48] private equity and private credit and [51:50] hedge funds, infrastructure, things like [51:51] that every two years. And at the end of [51:55] and, you know, as a courtesy, I I send [51:57] the sections to some of the people that [52:00] we partner with in terms of money [52:01] management. And I sent the private [52:04] credit section to somebody at [52:05] Blackstone. Uh, and um, this was last [52:10] November. And one of his many criticisms [52:14] of my private credit section, which was [52:16] the longest section in the piece because [52:17] of all the concerns we have, is that I [52:19] was underestimating [52:22] um the value proposition of their [52:24] software exposures, [52:25] >> right? [52:26] >> Underestimating. [52:26] >> Underestimating you and the rest of the [52:28] world is now underestimating. [52:30] >> That's how quickly the narrative [52:32] changed. Last November, they saw that as [52:34] a selling point and they also had, you [52:37] know, probably close to the highest [52:38] software exposure. [52:39] >> That's right. November, nobody was [52:40] really talking about the death of [52:41] Salesforce. [52:41] >> That's how quickly it changed because by [52:44] February, you know, that wasn't a [52:48] discussion he wanted to have. [52:49] >> Michael, you wrote, um, here's the [52:50] important point. According to JP Morgan, [52:52] Equity Research, the unit economics of [52:53] AS6, so we're talking about Google TPUs, [52:55] AWS, Trrenium, Microsoft, Maya, Meta, [52:58] NIA, [52:59] >> they're improving versus Nvidia. You [53:01] wrote that the hyperscaler is using [53:02] their own in-house chips support total [53:04] cost of ownership reductions of 30 to [53:05] 40%. [53:06] >> Right. [53:06] >> Okay. Um, is this why you think Nvidia [53:09] is stuck in the mud? Because the photo [53:10] pee of Nvidia is as low as it's been. I [53:12] think it's at 18 times forward. [53:13] >> It's 20 right now. [53:14] >> That's I don't think that's bullish. [53:15] That's not a good thing. I don't think [53:16] Nvidia getting cheaper is like awesome. [53:18] So, first of all, like the thing you [53:19] just read is is is a great advertisement [53:22] for the other thing of me saying this is [53:25] this is more complicated than other [53:26] technology revolutions because because [53:28] for the last few years we were, you [53:31] know, all told that Nvidia GPUs were [53:33] untouchable and that the efforts that [53:35] the other hyperscalers [53:37] were making to develop their own [53:40] alternatives were not ready for prime [53:43] time and that they were stuck [53:45] contributing forever to Nvidia's 70% [53:48] gross margin. That was wrong. And um all [53:52] those companies that you just mentioned [53:53] have been able to build their own [53:55] accelerators. Right. These are all fall [53:57] under the category of accelerators that [53:58] migrate compute away from the CPU and [54:01] they've and none of them have [54:04] successfully yet figured out how to or [54:07] even if they want to sell them broadly [54:10] to third parties because you'd need a [54:11] salesforce and you'd need customer [54:13] support and Nvidia has all that. you [54:15] need all the software. But they were [54:18] huge customers of Nvidia. So even if all [54:20] they do [54:21] >> is build their own tools to replace the [54:24] Nvidia stuff they used to buy, [54:26] >> that's big news. [54:27] >> And yet and yes, they and yes, Nvidia [54:29] has a trillion dollar backlog, but [54:31] that's not the point. I listened to Andy [54:33] Jasse on a on a call maybe last quarter [54:36] or the quarter before explain that if [54:39] Amazon's tranium business internal would [54:44] be valued the way that we value other [54:46] semiconductors business like the annual [54:49] revenue that that would be equivalent to [54:52] it would be one of the largest companies [54:54] in the world. [54:54] >> Yes. But but again right now it's it's [54:56] it's mostly internal to Amazon [54:58] >> for but couldn't that change on a dime? [55:00] It it it it could, but again [55:03] the cloud was internal to Amazon until [55:05] one day it wasn't. [55:06] >> But it it doesn't even need to to be a [55:08] problem for Nvidia, which is I think [55:10] what you're mentioning, which is like [55:11] these are some of Nvidia's biggest [55:13] customers saying, you know what, we [55:14] we'll buy we're going to buy some GPUs [55:16] but less than we used to because we're [55:18] building our own and and saving a lot of [55:20] money. [55:20] >> And the market is sussing that out [55:22] because it is not it is the the P the [55:23] Forward P is shrinking and it has been [55:25] for a long time. Yeah. And I think the [55:26] market is being quite sober in a lot of [55:28] different areas despite the fact that [55:31] there is a lot of insane behavior, [55:33] despite that there is a lot of leverage. [55:36] Josh and I were talking last week, the [55:37] amount of money in these double levered [55:40] ETFs. There's like $500 billion of [55:43] notional exposure. [55:44] >> And this is why you see the giant [55:46] whipsaws dayto day. [55:47] >> I spent some time with our derivative [55:49] guys. [55:49] >> Are you worried at all? [55:51] Well, I mean, I'm I spent a lot of time [55:54] with our derivatives guys understand I [55:56] had a I had a weird chart in the piece [55:59] that looked at the the contribution of [56:02] these leverage ETFs to changes in daily [56:04] market cap swings. [56:06] >> Yeah. [56:06] >> And [56:08] you know the chart nine, please, Daniel, [56:10] >> the way that they're hedged, right? [56:13] >> So, the bottom right, right? [56:14] >> The bottom right chart. [56:15] >> Unbelievable. So, what what is what are [56:16] we looking at here? [56:17] >> Oh, boy. So, [laughter] [56:19] >> oh boy. [56:20] >> So, [clears throat] the way that these [56:23] providers hedge these exposures is they [56:26] have to buy on the way up and they have [56:28] to sell on the way down. Essentially, [56:30] imagine a a market neutral fund that's [56:32] got to go home flat at the end of every [56:33] day. So, if they've offered these these [56:36] these leveraged ETFs, they have to [56:39] constantly be chasing the market up and [56:41] down. And as you can see, just this [56:44] year, the amount of money that's been [56:46] pouring into this stuff has has [56:48] increased a lot. And you know, and these [56:51] that just means, like you said, you're [56:53] going to get a lot of [56:55] 2:30 p.m. to 4:00 p.m. noise. And uh [57:00] it's just I I think it's going to be [57:02] kind of spooky for the average investor. [57:03] >> If these if these stocks peak and then [57:05] trade flat or trade in like a 10 or 15% [57:08] draw down from their highs, these are [57:09] all dead [57:10] >> for extended period. This activity [57:12] though will go away because it's not fun [57:14] anymore. [57:14] >> And also I mean if they're flat it's [57:17] also a problem for the construction of [57:19] the product because [57:21] you have negative embedded leverage, [57:23] right? And so you've got the cost. [57:25] >> What I'm saying though, dollar value [57:26] will come out of this game. They'll go [57:28] on couchy. They'll do something [57:30] different. [57:30] >> My my favorite indicator on this page [57:34] for for better for worse is the the [57:36] chart on the Cosby. So the margin loans [57:39] on the Cosby. And all I can tell you is, [57:41] and I'm remembering back to 2008 and [57:45] 1997, [57:46] Korean retail is basically the, you [57:50] know, the the scariest market indicator [57:52] cuz by the time those guys start to pile [57:54] into stuff, you know, you're you're [57:56] really really really late in the game. [57:58] >> This looks This is piling in. [58:00] >> Yeah. Yeah. [58:00] >> But in Yes. And in fairness, there's a [58:02] reason why they're piling in. Samsung [58:04] and SK. [58:05] >> Yeah. [58:06] >> The the growth there is otherworldly. It [58:08] it is, but again, they're the they're [58:11] the last train on the caboose, and we [58:14] we'll need over the next six months a [58:17] sharp rebound [58:19] in in in the hyperscaler [58:22] projected cash flows and AI adoption in [58:25] order in order for this all to settle [58:27] out. Okay. [58:27] >> So, when when you see Amazon sell $30 [58:30] billion in bonds or you see um [58:33] >> that doesn't worry me yet. That that [58:35] still puts [58:35] >> the numbers are not big enough. that [58:37] still puts their debt to cash flow [58:38] ratio. I [58:39] >> bet Alphabet selling stock after 10 [58:41] years of buybacks. That's different. [58:43] >> Yeah, that was different. [58:45] >> Uh e e equity fundraising for for capex [58:49] is not [58:50] >> I don't think anybody [58:51] >> Well, I mean it tells you we looked at [58:52] the chart already. They're not [58:53] generating the kind of internally [58:54] generated cash flow where they could [58:56] just write some internal checks to do [58:57] that, [58:58] >> right? [58:58] >> Um so I wanted to get your opinion on on [59:01] SpaceX. I thought the lockup, what they [59:04] were doing, why not? I think the way [59:07] that it's done right now is so gified [59:10] and so predictably awful. It doesn't [59:12] work. This the typical six-month lockup. [59:15] You see shares fall in, they sort of [59:17] settle out. Um I don't know that I love [59:19] that they only release 3% of the company [59:20] or sold 3% of the company. I think that [59:22] like, you know, they did that very [59:23] intentionally. I don't really love that [59:24] part of it. But to have like this [59:26] staggered lock up, some uh price points [59:30] where more shares could be locked up, I [59:31] I didn't hate it. [59:33] >> Yeah. The question is like, you know, [59:35] direct placements as an alternative to [59:37] IPOs, like let's let's see how it works. [59:39] I mean, at the end of the day, by the [59:43] end of this year, we're going to go from [59:46] something like 3% to something like 50% [59:48] of the shares being part of the free [59:50] float. [59:51] >> I don't know. I don't know if you there [59:52] is a lot of analyst support that that [59:54] has come in this week predictably [59:57] obviously. I don't know if you saw the [59:58] FT Alphavville by recommendation. Uh [60:01] >> stop. [60:03] >> FT Alphavville initiated coverage today [60:05] on SpaceX with a 12-month price [60:08] objective of infinity. [60:10] >> They don't do that. Stop it. [60:12] >> Tell them Tell them you saw it. Tell [60:14] them you saw it. You don't have to [60:15] comment on it. [60:16] >> Do you want I saw it. I I don't want I [60:19] don't know. They literally think I'm [60:21] joking. I'm not joking. [60:22] >> I don't want to comment. [60:23] >> No. Okay. There'll be no comment on [60:24] that. Uh if you if you if you look at [60:28] though the way that SpaceX placed and [60:31] priced and the activity since I know [60:34] it's been only a few days, [60:36] there were a lot of really dire [60:38] predictions out there about how it was [60:40] going to screw up the indices. People [60:43] going to maybe people dump stocks, but [60:45] we can't really see it on a chart. [60:46] >> It's in the cues. There's no evidence [60:48] yet that anything super funky happened [60:50] as a result of SpaceX's coming public [60:52] and it being included in the index. At [60:54] least not yet. [60:54] >> I'm going to say so like so far the [60:58] worst prediction. [60:58] >> So far it's fine. And I and and I like [61:00] the fact that S&P stuck to its guns and [61:04] did not make an exception about you know [61:07] companies having to be profitable before [61:09] they get and put into the index. [61:10] >> You you like that they did. [61:12] >> Yeah. Okay. Okay, [61:12] >> somebody has to have some profitability [61:14] standard. Or else you end up with a [61:15] Russell 2000, which is a pile of dog [61:17] [ __ ] companies that don't make money, [61:19] >> right? The NASDAQ put it in because it [61:20] trades on the NASDAQ. So, if you're [61:22] going to let this thing trade, you have [61:24] to have a good enough opinion. [61:25] >> I think that I wouldn't be surprised to [61:27] find out the S&P and NASDAQ and a bunch [61:30] of people all kind of discussed like how [61:32] can we split how can we split the pie [61:34] here in a way that's reasonable. Okay, [61:36] you guys put it in your benchmark. We're [61:38] not until the pro, you know, like it [61:40] wasn't an all or nothing thing. And so I [61:42] think it's a reasonable compromise the [61:43] way the indices chopped this thing up. [61:45] >> How do you feel about the health of the [61:46] bull market right now? [61:48] [sighs and gasps] [61:48] >> Um [61:49] >> let me let me let me lead you to water. [61:51] So Josh and I were talking last week [61:53] about if you are just looking at the [61:54] market inside the market of things that [61:56] you would want to see participating in a [61:57] bull market to give you confidence that [61:59] things are going pretty okay. [62:01] >> Yeah. [62:01] >> Regional banks, transports, industrials, [62:05] small caps, micro caps, um [62:08] >> Wall Street banks. [62:09] >> Wall Street banks. City Croup on fire. [62:11] Like there is a lot there is a lot to [62:13] really like about the current state of [62:15] the market. I'm not saying it's going to [62:16] happen for continue forever, but just [62:18] today diagnosing the health of the [62:20] market looks pretty good. [62:21] >> The internals look okay. You know, um [62:24] obviously we've got some [62:27] strange valuation shifting going on in [62:30] terms of what's going on with the [62:31] hyperscalers and the semiconductor [62:33] companies. But the market, you know, the [62:36] market breath numbers are not great. [62:38] Transports look great though with the [62:39] Dow and then away from the stocks. Uh [62:43] retail sales were way better than [62:45] expected. [62:46] >> Yeah. Some of [62:47] >> loan growth at the banks. Like these are [62:49] all signs of a healthy environment. [62:51] >> When the the the thing that we look at [62:53] the most [62:54] >> is the RA is is in the PMI indices. It's [62:57] the ratio of new orders to inventories [63:00] of of all the things that we follow. We [63:03] follow hundreds of them. That [snorts] [63:04] one has the closest relationship with [63:06] the performance of the stock market. [63:08] >> What sort of orders are they measuring? [63:09] Because I remember you spoke about that [63:10] once before [63:11] >> and new manufacturing orders. [63:13] >> You could also look at news orders and [63:14] services, but I don't like it. It's it's [63:16] unclear what exactly they're doing. But [63:18] new manufacturing orders is very clear. [63:20] And um and remember we talked about the [63:22] S&P as a production index, not a [63:24] consumption index, right? So new [63:25] manufacturing orders relative to [63:28] inventory accumulation. It's and and we [63:30] have this link that has this Trump [63:31] tracker that has all the charts on the [63:33] economy that we follow. People can look [63:34] at it and um and that that looks okay. [63:38] And a lot of the contemporaneous [63:40] coincident leading indicators and [63:41] forward leading indators look okay. Um [63:44] consumer part a little shaky. Um rising [63:48] credit card delinquencies, rising auto [63:50] delinquencies, but again, you know, [63:53] those aren't huge drivers of corporate [63:55] profits. Do you uh think the election is [63:58] going to have an impact on the stock [63:59] market in uh the second half of the year [64:01] or the fourth quarter of the year? [64:03] >> It usually doesn't. Um [64:06] >> it's very noisy, but I can't remember a [64:08] midterms that made me feel differently [64:10] about what I was invested in. [64:12] >> Yeah. You know, it's kind of like, you [64:15] know, you know how there are certain [64:17] things you just don't spend time on [64:18] because it's a waste of time and you you [64:20] either need to do other work or rest. [64:22] That's how I feel about the uh [64:24] >> You're not getting questions yet about [64:26] >> Well, that's how I feel about the State [64:27] of the Union address. Like it it it's [64:29] meaningless in the scheme of things. And [64:30] I kind of the midterm elections, you [64:33] know, rarely have have substantial [64:36] policy implications for investors. Now, [64:38] you know, Democrats take the House and [64:40] the Senate. What happens to the Trump [64:42] agenda? Do they impeach him? Um again, I [64:45] don't know. Um but we'll see. I mean, we [64:49] we if you want, we can spend a couple [64:51] minutes talking about redistricting and [64:52] things like that, but [64:54] >> Okay. Do you worry about whether or not [64:57] Anthropic and Open AAI will be able to [65:00] actually go public this year? And if [65:02] they have to postpone into next year, [65:04] could that potentially be looked at as a [65:06] threat to the earnings expectations of [65:08] all of the suppliers to the AI [65:10] ecosystem? Like, is that a thing that [65:12] you're concerned about or not really? [65:15] So, a lot of the capex is supposed to [65:17] come from these companies, right? [65:19] >> And ostensibly, they they're going to [65:20] need liquidity to hit these commitments [65:23] that, [65:23] >> right, [65:24] >> everyone's got baked into their earnings [65:25] expectations. [65:26] >> Yeah. [65:28] Well, um, [65:31] you know, S&P buybacks are still running [65:33] at a pretty decent clip and there's a [65:36] lot of cash M&A [clears throat] taking [65:38] place. So, cash has been around 70% of [65:40] of M&A compensation this year. So, I'm [65:43] not worried that there's like quote [65:44] unquote not enough cash to buy these new [65:47] IPOs. I don't think that's an issue, [65:48] particularly when they're only selling [65:50] 3% of the float, [65:51] >> right? [65:51] >> So, um it makes it very easy when [65:54] there's no stock for sale. [65:55] >> So, but I'm as a general principle, like [65:59] let's take away some of the valuations [66:01] and the noise since the IPO. [66:03] I I think [66:06] if I had to rank the business the [66:08] underlying business uncertainty, [66:11] I would say OpenAI the most, then [66:14] anthropic, than SpaceX. [66:15] >> I think the market is I don't think the [66:16] market's going to like Open AI. [66:19] >> I think it has more business uncertainty [66:22] than the other two, right? I mean, and [66:23] and you know, everybody everybody again, [66:25] everybody's going to have their own [66:26] ranking, but that's mine. [66:28] >> Why why do they have more business [66:29] uncertainty? [66:31] I think when you look at the [66:33] concentration of how open I open AAI [66:35] makes money um there's not a lot of [66:38] there's not a lot of advertising [66:40] revenue. It's a subscription service. [66:42] >> Not as much enterprise. [66:43] >> Yeah. I mean [66:44] >> it's much more consumerf free [66:46] >> and th those are the kind of things [66:48] where where sentiment can shift. I just [66:50] I don't it doesn't look to me to be the [66:52] the the dynamics don't look quite the [66:54] same as as the other two companies. [66:57] >> Yeah. They sort of have a flavor of like [66:59] AOL in the early internet to them where [67:02] it was like the first thing that [67:03] everyone used but nobody stopped there. [67:07] >> Yeah. I mean the network effect is is is [67:09] pretty powerful but yeah [67:10] >> you know anthropic showed how quickly [67:12] those kind of things can change. By the [67:14] way, um, one of my summer projects is [67:18] um, our COO has asked me, so we're JP [67:22] Morgan's a member of Project Last Wing, [67:24] which is the the [67:25] >> the cyber security [67:26] >> Yeah. the 20 or 30 companies that were [67:28] given mythos. [67:29] >> And so, um, the COO, uh, asked me to [67:34] partner with our cyber security team to [67:36] write a piece on what we've learned so [67:38] far from Project Glass Wing. And, um, [67:40] what's coming and, uh, so I'm I'm in the [67:44] midst of working I'd probably publish [67:46] that by the end of the month. Um, you [67:48] know, that's an eye openener. The the [67:50] name of the piece is called [67:51] Patchmageddon, [67:52] [snorts] [67:53] >> right? Because you know, you know the [67:56] story. Um, [67:57] >> Bruce Willis went to space. Of course. [67:59] >> Well, no, they whatever that is, I [68:02] didn't see it. [68:03] >> They they said before we release this to [68:05] the public, we should let these [68:06] companies know how many things need [68:09] patching in their existing [68:11] >> and and the the pace of of confirmed [68:14] vulnerabilities is going up like this [68:15] and the pace of patching is down here. [68:17] It's just and and there's also physical [68:19] infrastructure issue with programmable [68:20] logic controllers. you this is going to [68:24] be a carnival for bad actors. [68:26] >> Okay? [68:26] >> So, state sponsored and non-state [68:29] sponsored bad actors are going to have a [68:31] field day for some period of time [68:35] um using these models to discover [68:37] bizarre hidden vulnerabilities that [68:40] nobody knew existed. [68:42] >> Yeah. And um so when when you look at [68:44] the vulnerabilities that these programs [68:46] find, it's like well you know if you log [68:48] in this way and then you do this other [68:51] strange thing and then you hit the [68:52] following three keystrokes all of a [68:53] sudden you can execute you know a [68:56] corporate takeover and like and it they [68:58] find these things that you would that no [68:59] human would ever have tried to do. [69:01] That's why they're so good at drug [69:02] development because they can work [69:04] through all these permutations that [69:05] nobody would and combinations of [69:07] molecules that nobody else would ever [69:08] think to do and then all of a sudden [69:10] they find like they did recently an [69:12] antibiotic uh a a a molecule that can [69:15] combat this antibiotic resistant MRS. [69:18] >> It really it really does feel like it's [69:20] an alien in our midst that does not [69:22] think like humans think. It arrives at [69:25] answers that are palatable to the human [69:29] that the answer lands on. But neural [69:31] networks don't know how it got there. [69:33] >> As a matter of fact, um a lot of banks [69:36] started using neural networks two three [69:38] years ago to um approve or deny uh uh [69:44] debit card and credit card uh [69:46] expenditures on the spot. And um and [69:49] because they weren't able to describe [69:51] how those neural networks actually [69:53] worked, they had to work, people [69:55] complained to regulators and the banks [69:57] have had to uh to overlay these new [70:00] tools that can try to take what the [70:02] neural network did and translate it into [70:04] English, which it doesn't always do such [70:05] a great job at, [70:06] >> right? [70:07] >> Cuz because you might get the the best [70:09] outcome a 100 times, but on the 101st [70:12] time, you get a very bad outcome. And if [70:14] you can't explain where it came from or [70:16] why, it's problematic. [70:18] >> I I don't think it's going to amount to [70:19] like a marketwide thing, but you know, [70:22] buckle up because individual small, [70:25] midcap, and large companies are going to [70:27] end up uh in the headlines from time to [70:30] time because they got breached and [70:33] attacked and it'll they'll have to kind [70:35] of sort out what they do about. Can [70:36] >> I can I make an observation? Yes. Tell [70:37] me if you agree with this. I feel like [70:40] you're in all your glory right now with [70:42] all these crossurrens and all these like [70:44] new technological things. I feel like [70:46] you're having a lot of fun trying to [70:47] solve these puzzles. Do I have that [70:49] right? [applause] [70:50] >> Do you do you agree? [70:52] >> No. You [70:53] >> You're not enjoying it? [70:54] >> No. I'm 64 years old. [70:56] >> Yeah, but you hard like my brain is [70:58] having to work so much harder per unit [71:01] of of output. I feel I I feel like [71:05] you're I feel like you're like perfectly [71:07] positioned amidst all of these things [71:09] that you've been studying your whole [71:11] life and where they overlap and uh all [71:14] the lessons that [71:15] >> you might not feel this way but we do. [71:17] >> We're getting a lot from each time you [71:19] publish something immediately new [71:21] semblance dropped [71:22] >> and we read it. [71:23] >> That's how we feel when when you [71:24] publish. [71:24] >> That's I remember my kids used to talk [71:26] that way when there was a new Drake [71:28] album. They would say, "Oh, no wait, not [71:29] Drake. The other guy." Oh, you see now [71:32] I'm not even doing it. [71:32] >> Well, that's you. You're the Drake of [71:34] Frank. [71:34] >> Frank Ocean. [71:35] >> Yeah. New Frank Ocean. [71:37] >> They used to say something dropped and [71:38] they would be all excited about it. [71:40] >> Yes. Uh well, we feel the same way. We [71:42] feel that way about your research and uh [71:44] you may the one thing is you may get an [71:47] email every once in a while that's like [71:48] not a fan. [71:50] >> That's okay. Goes with the territory. By [71:51] the way, Jamie is great at like who [71:54] cares, you know? Yeah, you got everyone [71:57] has their own [71:57] >> opin there's a million people who are [71:59] reading you that don't send an email but [72:01] they love what you're doing. You just [72:03] won't hear from them as often as you'll [72:06] hear from the one person that's got an [72:07] issue with one thing that you said. [72:09] >> So don't overindex to that would would [72:11] be would be as a veteran of of getting [72:14] feedback for things I say in public. [72:16] >> Yeah. Although can I tell you a lesson [72:17] that I learned? [72:18] >> Yeah, please. [72:18] >> And you know you you you learn things at [72:20] different points in your career. I um [72:24] [snorts] I wrote I write the energy [72:25] paper, you know, it's I probably spend [72:27] three or four months on it. It's the [72:28] it's the thing I work the most on each [72:30] year. And this one was called Fighting [72:33] Words because of all the debates and [72:35] things like that uh that people have [72:37] about energy. And the cover was a [72:39] three-way shootout um from the movie uh [72:41] The Good, the Bad, and the Ugly. And it [72:43] was a three-way shootout between [72:44] advocates of nuclear, advocates of [72:46] renewables, and advocates of fossil [72:47] fuels. you find all three of those [72:49] represented in society and within the [72:51] administration itself. And [72:54] um I I I had a couple of occasions where [72:57] a certain client was um very voseiferous [73:02] with me about their concerns about the [73:05] way that I was evaluating the energy [73:07] markets and um uh and you know and [73:11] confronting me about it. And so I wrote [73:14] about that in the introduction to the [73:16] energy paper. [73:18] and um anonymously so that nobody would [73:21] know who this person was but they knew [73:23] who it was [73:24] >> and they were very very upset. [73:27] >> They were upset. Why? Because you used [73:29] their argument. [73:30] >> Oh, I I used I used that interaction [73:34] which wasn't the the easiest one to deal [73:38] with. I used that interaction as a [73:40] metaphor for the for the fierce debates [73:42] people are having about energy. And I I [73:44] probably used certain words or phrases [73:46] that to describe those interactions um [73:49] which which weren't really super [73:51] wellreceived. [73:51] >> So what's the lesson you learned? Don't [73:53] do that. [73:53] >> Oh, I learned lessons. [73:55] >> Okay. Um [73:55] >> what did you learn? [73:56] >> Um that there's there's a line not to [74:00] cross there. And um and uh I got I got [74:04] some phone I got some calls from Mary, [74:06] >> right? [74:07] >> Uh uh from Mary Erdos, my boss, about [74:09] that particular one. um because you [74:12] don't know, you know, exactly who people [74:15] are. And um but it's it it's ending up [74:18] in a good place because what I've [74:20] decided to do is I'm going to this is a [74:23] first for me in my entire career. in [74:25] next year's energy paper, I'm gonna [74:27] allow that person to select the policy [74:31] energy expert of their choice to write [74:34] an unedited, unexpated [74:37] um section in the energy paper on why [74:40] they disagree with me and on what. [74:42] >> So that point of view will make its way [74:44] into the research. It will. And you [74:45] won't have you won't have to anonymize [74:47] the person. You could just use [74:48] >> So it wasn't kind of planned that way, [74:50] but you know, I [74:52] >> we all do what we got to do. [74:53] >> Yeah. So, you know, you learn lesson. [74:55] You learn different lessons at this at [74:56] different points in your career, and I [74:58] learned one of them this year. [75:00] >> All right. Well, fair enough. Well, [75:01] listen, we really appreciate you coming [75:03] by, walking us through some of these [75:05] charts. I want to tell people where they [75:07] can find more from Michael Semolist. Um, [75:09] >> I generally post [75:11] >> Yeah. [75:11] >> almost everything. [75:14] That's right. Okay. So, f So, you want [75:15] to follow Michael on LinkedIn, of [75:18] course. And then you do you drop videos [75:20] when these reports come out. [75:21] >> I tell Michael I love the video. Yeah, [75:23] the digest. [75:24] >> Each each uh each eye in the market is [75:26] accompanied by a like video podcast [75:30] >> that I record from my home. So, [75:34] >> okay. [75:34] >> Um [75:35] >> All right. So, if you [75:36] >> And remember, you can also follow my [75:38] fishing adventures. I do [75:40] >> on Instagram. [75:42] >> Oh, boy. What do we have here? Wait, [75:44] what are we doing? That better not be a [75:46] >> No firecrackers. [75:47] >> No firecrackers. [75:48] >> Wait, wait, wait. Wait. Is she serious? [75:52] Rob, you have responsibility. [75:54] >> SPARK. [75:55] >> Happy 250 everybody. [76:01] [music] [76:02] >> Very nice. [laughter] Well done. [76:03] >> This is a unique program. [76:07] >> Nicole, wait till you see your footnote. [76:09] [laughter] [76:11] >> Holden Cfield, here we come. [76:13] >> All right. Thank Thank you to Michael [76:14] Seis, ladies and gentlemen. Thank you so [76:16] much for watching. Thank you for [76:17] listening. Happy 250th America. and uh [76:20] we'll talk to you soon. We'll see you [76:22] again. Thanks again. [76:28] [applause]