Anthropic Revenue Surge Re-Ignites AI CapEx Boom As Cerebras Debuts

Anthropic营收激增且Cerebras上市，重燃AI资本支出热潮 随着大规模生成式AI基础设施支出推动商业模式转向资本密集型实体资产，科技巨头面临估值遭受结构性压缩、沦为低估值倍数公用事业企业的风险。 专家小组深入分析了推动AI交易的巨额资本支出，并将Anthropic迅猛的营收规模扩张与Cerebras的IPO，同历史资本周期及过度投资风险进行了权衡探讨。他们辩论了传统软件股是否正面临结构性颠覆，并剖析了超大规模服务商的资产负债表为何逐渐趋同于实体公用事业企业。Kai Wu分享了在现代量化框架下评估无形资产的见解，而Ben Carlson则审视了历史市场回撤与长期的风险回报动态。 Anthropic 年化净金额留存率（NDR）突破 500%，年化营收在短短一个季度内从 90 亿美元跃升至 300 亿美元，模型采用率呈现前所未有的指数级爆发。 (12:17) AI 芯片公司 Cerebras Systems 以近 1000 亿美元市值在纳斯达克挂牌上市并募资 55 亿美元，创下 2020 年 Snowflake 以来美股最大规模的科技 IPO。 (14:55) 英伟达在 7 个交易日内市值暴涨 9000 亿美元，当前对应远期市盈率约为 27 倍，未来 12 个月预期净利润高达 2230 亿美元。 (19:23) 美股市场呈现显著的 K 型分化：传统软件股遭遇连续 25 周走低的罕见抛售，而半导体板块在标普 500 指数中的权重已急剧膨胀至 23%。 (34:08) 分析人士警惕历史资本周期重演，随着科技巨头向资本密集型的数据中心与电力基建倾斜，其商业模式或将向回报率较低的公用事业股发生估值下修。 (48:51)

While surging corporate AI expenditures and hyper-growth at frontier labs continue to drive market optimism, panelists debate whether hyperscalers are entering an overbuilt capital cycle where downstream adopters capture the economic surplus while infrastructure operators face lower valuation multiples. Hyper-Growth Financing and Frontier Infrastructure Spend (14:55) Anthropic's financial metrics reflect unprecedented enterprise adoption, with leadership citing an annualized net dollar retention rate exceeding 500% and run-rate revenues expanding from $9 billion to $30 billion in a single quarter, toward a target of $50 billion. This capital influx was echoed by the Cerebras Systems initial public offering on the Nasdaq, which raised $5.5 billion at a valuation approaching $100 billion against roughly $1 billion in annualized recurring revenue. The offering was 20 times oversubscribed, despite notable customer concentration that includes an initial 750-megawatt inference agreement with OpenAI scalable to 2 gigawatts by 2030. Sparkline Capital CIO Kai Wu notes that this commercial momentum occurs while frontier labs subsidize token access below true production costs to secure dominant market share. Consequently, current token pricing does not yet provide an accurate economic signal, leaving open the question of whether enterprise adopters can generate sustainable returns on their software investments once subsidized pricing normalizes. Fundamental Earnings Contrast with the Dot-Com Bubble (40:12) Panelists reject direct comparisons between the present rally and the 1999 dot-com bubble by examining underlying index profitability. At the 1999 market peak, 66% of the Nasdaq index traded above 60 times earnings or was entirely unprofitable, with 10.5% of constituents generating negative profit margins. Today, roughly 60% of the index trades between 20 and 40 times forward earnings, and 70% operates with margins above 25%, underpinned by a 20-year compound annual earnings growth rate of 14%. Even in high-momentum segments, valuation multiples remain tied to earnings delivery. The VanEck Semiconductor ETF trades near 27 times forward earnings despite a 220% advance from its cycle low, while Nvidia's market capitalization is supported by an estimated $223 billion in projected 12-month net income. Hosts Michael Batnick and Ben Carlson conclude that while regular corrections remain possible, the structural profitability of current market leaders makes an 80% post-1999 style crash improbable. Capital Cycle Overbuilding and Utility-Style Re-Rating Drawing on financial historian Edward Chancellor’s capital cycle theory, Kai Wu observes that infrastructure booms historically lead to overinvestment, excess capacity, and sharp demand air pockets. Hyperscalers such as Alphabet, Meta, and Microsoft find themselves in a game-theoretic prisoner's dilemma where failure to match rival capital spending carries perceived existential risk, driving massive expenditures into data centers and proprietary hardware. Wu argues that transforming from asset-light software businesses into capital-intensive physical infrastructure operators may cause these hyperscalers to be re-rated downward toward utility-like multiples. Josh Brown compares this to the 1990s telecom buildout and 19th-century railroads, where network builders suffered prolonged stock stagnation or bankruptcy while downstream layers, such as Apple or rail-reliant manufacturers, captured the enduring equity value. Wu also notes an accounting distortion: hyperscalers capitalize and depreciate hardware over roughly five years, temporarily shielding net income while hardware vendors book immediate revenue. Software Valuation Compression and Intangible Asset Measurement (34:08) Market breadth shows a notable internal divergence: the S&P 500 has set record highs while over 5% of its members simultaneously touched 52-week lows, a rare occurrence previously recorded only in 1929, 1973, and 1999. Public enterprise software firms have experienced prolonged declines, with Adobe trading at 9.5 to 10 times price-to-earnings despite record free cash flows, reflecting market fears that AI-generated code will erode software pricing power. Wu contends that enterprise relationships and distribution channels, rather than raw code, represent the true competitive moats for incumbent vendors. Modern quantitative investing, Wu explains, must shift away from obsolete book-value metrics that ignore intangible capital. By deploying natural language processing across unstructured data sources such as patent filings, earnings transcripts, and search momentum, quantitative managers can systematically price intangible assets like research productivity and brand equity. While the high margin profile of modern index leaders limits the probability of a drawn-out structural depression, passive investors face extreme concentration risk, with AI and semiconductor themes representing an estimated 50% to 60% of broader index movements. An unresolved question is whether physical bottlenecks like electrical grid capacity and chip supply will act as a stabilizing governor, giving corporate balance sheets and labor markets time to adapt before overinvestment leads to an infrastructure air pocket.

本期节目深入探讨了人工智能驱动下的美股繁荣是否已触及极限。与会嘉宾认为，尽管当前纳斯达克成分股的盈利质量显著优于1999年互联网泡沫时期，但科技巨头因“囚徒困境”展开的大规模资本开支正迫使其由轻资产向重资产公用事业模式转变，未来可能重演电信与铁路热潮中基础设施投资者回报受限、而下游应用获益的历史悖论。 盈利支撑下的AI热潮与1999年对比 (14:55) 本轮AI交易在经历前期震荡后再度升温，底层企业呈现出惊人的增长速度。Anthropic年化运营收入在单个季度内从90亿美元跃升至300亿美元，内部超90%的代码由Claude Code编写，净金额留存率超过500%；Cerebras以约1000亿美元市值上市，创下2020年Snowflake以来最大规模美股科技IPO，超额认购达20倍；英伟达更在短短7个交易日内市值激增9000亿美元。然而，Michael Batnick与Ben Carlson指出，当前半导体行业市盈率约27倍，纳斯达克100指数过去20年间维持了14%的年化盈利增速，目前的涨势有着扎实的盈利基础支撑。与1999年互联网泡沫顶峰时三分之二成分股市盈率超60倍或处于亏损状态截然不同，如今纳斯达克100指数中70%的公司利润率超过25%，几乎不存在无营收的投机性企业，因此重现当年80%毁灭性暴跌的概率极低。 资本周期陷阱与巨头重资产转型 (29:26) 尽管短期业绩强劲，科技巨头的激进投资正引发资本周期的经典警报。Kai Wu引用金融史学者Edward Chancellor的资本周期理论指出，历史上除形成垄断的电话网络外，几乎所有基础设施热潮（运河、铁路、电力与互联网）都因供给过剩和需求断层而走向崩盘。当前前沿AI实验室以低于经济成本的价格补贴代币（Token）以抢夺市场份额，而科技企业CEO受制于博弈论中的囚徒困境，因担忧在通用人工智能竞赛中出局而被迫过度投资。更为严峻的挑战在于商业模式的质变：过去十五年，Alphabet、微软和Meta凭借无形资产和网络效应享有超高的已投资资本回报率（ROIC）与轻资产估值溢价；而如今为了建设庞大的数据中心，巨头们正转变为重资产的基础设施运营商。Kai Wu强调，科技七巨头（Magnificent Seven）的基准风险并非破产，而是估值中枢向重资产公用事业股靠拢的持续下调；同时Chamath Palihapitiya提及的5年期折旧滞后会计处理，也在短期内掩盖了巨额硬件采购对当期真实利润的侵蚀。 基础设施悖论与电信铁路历史镜鉴 Josh Brown与Kai Wu将当前的AI军备竞赛与1990年代电信基建及19世纪铁路繁荣进行了深度类比。在光纤与移动网络建设期间，AT&T和Verizon等电信巨头倾注了数千亿美元资本，但其股价历经数十年却基本处于原地踏步状态，回报率退化至债券水平；电信基础设施释放的真正股权红利最终被苹果等在其网络上构建生态的应用层所捕获。同样，Azeem Azhar整理的19世纪美国铁路繁荣数据表明，铁路建设开支数次超越运营商利润并引发倒闭潮，铁路完工后利润率始终平平，但整体经济GDP以及利用铁路降本的实体产业（如家具制造商）却获得了巨大飞跃。嘉宾们由此推论，本轮超大规模企业的资本开支最终可能会将底层芯片算力与大模型层严重商品化，超额利润更有可能流向终端消费者以及类似Palantir等专注落地实施的企业级应用层软件，而非承担沉重基建负担的科技巨头自身。 软件股护城河危机与K型市场分化 (34:08) 指数的整体繁荣掩盖了市场内部剧烈的K型撕裂。标普500指数创历史新高的同时，超过5%的成分股创下52周新低，这种背离在历史上仅出现于1929年7月、1973年1月和1999年12月。当前半导体板块已占到标普500总市值的23%，而传统上市软件股却遭遇长达数十周的持续下挫，Adobe等公司的市盈率甚至跌落至9.5至10倍的历史低位。Josh Brown提出，随着AI使代码编写成本趋近于零，市场正将许多传统软件公司视同失去壁垒的“传统报纸”；Kai Wu则认为软件企业的护城河在于客户分销网络与关系，但随着前沿AI机构联合咨询公司部署前线工程团队直扑企业端，传统软件厂商正面临前所未有的估值压制。此外，电网负荷上限与芯片短缺等物理瓶颈在客观上扮演了“调节器”的角色，适度放缓了AI全面部署的节奏，避免了短期内潜在的大规模失业对宏观经济造成骤然冲击。 无形资产重估与现代市场周期的压缩 (59:56) 投资范式与周期节奏同样在经历技术驱动的深刻重构。Kai Wu结合量化投资经验指出，Fama-French价值因子在1990年代后的钝化并非价值投资失效，而是传统会计体系无法准确度量软件、研发、专利和品牌等无形资产的真实价值；现代量化管理正借助自然语言处理（NLP）解析专利、财报乃至社交网络非结构化数据，以动态衡量企业的无形护城河。在宏观周期方面，Ben Carlson依据新书《风险与回报》的数据分析称，在即时信息传播与宏观政策干预下，类似大萧条长达数年的阴跌式熊市已被大幅削弱，现代市场更倾向于在1至3个月内极速完成超20%的暴力回调并快速复苏。针对市场对日本股市“失去三十年”的担忧，Carlson指出其1990至2024年年化1.5%的低迷源自前二十年高达22%的极端前置回报，若拉长至60年维度，年化复合回报仍接近9%，印证了只要坚持多元化与长期投资，权益资产能够穿越周期消化绝大多数危机。 与会嘉宾达成共识，当前的纳斯达克市场依托真实的盈利爆发，并不等同于1999年由空心投机主导的泡沫破裂前夕；然而，巨头间因防御性竞争而发动的AI资本开支竞赛，正在改变科技板块原有的轻资产高回报结构，带来估值中枢向公用事业下移的实质风险。未来市场的核心博弈不在于否定AI的技术价值，而在于辨识资本能否在基础设施商品化与软件护城河重塑的过程中，找到真正具有留存价值的利润分配层。

## Background 该视频由 YouTube 频道 The Compound 于 2026 年 6 月 13 日发布。根据官方简介，本期为《The Compound and Friends》播客第 242 期，由 Michael Batnick 与 Downtown Josh Brown 共同主持，特邀嘉宾 Ben Carlson 与 Sparkline Capital 创始人 Kai Wu 就 AI 繁荣、英伟达暴涨、Anthropic 的惊人增长、Cerebras IPO、软件股惨遭重挫以及科技巨头是否沦为新公共事业等话题展开了全面探讨。 ## TL;DR 节目围绕由万亿美元 AI 基础设施资本支出驱动的美股狂欢展开剖析，深入探讨了英伟达超常市值膨胀、Anthropic 爆炸式商业化变现与 Cerebras 抢购狂潮背后的资本周期规律。嘉宾们对比了当下的科技牛市与 1999 年科网泡沫的本质差异，指出虽然当前指数有扎实盈利支撑，但软件与芯片板块的极端背离、被动指数对科技巨头的过度集中以及庞大折旧对长期资本回报率的侵蚀均蕴含重大下行风险。此外，Kai Wu 分享了利用大语言模型量化品牌、专利等无形资产的前沿方法论，Ben Carlson 则结合其新书《Risk and Reward》拆解了极端回撤周期的压缩演变与长期乐观主义逻辑。 ## Atmosphere & style 整场节目在专业硬核与幽默风趣之间保持了绝妙的平衡，开场充满纽约金融圈特有的调侃与闲聊，随后无缝切换至犀利的资本周期推演与财报深度透视。主持人和嘉宾之间化学反应十足，在相互抛梗打趣的同时，不断用翔实的历史数据和冷峻的量化框架穿透当前的狂热叙事。收听体验既像置身于华尔街资深老友的闭门派对般轻松畅快，又带来深刻反思资本周期残酷现实的智力刺激与清醒感。 ## Bulletin - [14:26](https://www.youtube.com/watch?v=n889nI8sR84&t=866s) Anthropic 商业化数据呈现爆炸式增长，净美元留存率（NDR）年化超 500%，季度营收运行率在一个季度内从 90 亿美元跃升至 300 亿美元并有望冲上 500 亿美元，正驱动下游企业 IT 预算的大规模重构。 - [16:26](https://www.youtube.com/watch?v=n889nI8sR84&t=986s) 美股芯片新贵 Cerebras 拟募资 55 亿美元挂牌上市，凭借与 OpenAI 锁定的推理算力协议获得 20 倍超额认购，开盘估值逼近千亿美元，市销率高达近百倍。 - [34:10](https://www.youtube.com/watch?v=n889nI8sR84&t=2050s) 标普 500 指数在创历史新高的同时有超过 5% 的成分股跌至 52 周新低，历史上该极端分化仅在 1929 年、1973 年及 1999 年见顶前夕出现过，显现严重的“K 型经济”割裂特征。 - [48:51](https://www.youtube.com/watch?v=n889nI8sR84&t=2931s) 科技巨头陷入“囚徒困境”式的 AI 军备竞赛，由轻资产软件模式转向高额硬件资本开支，未来恐因沉重的折旧压力被市场作为类似公用事业股重新定价（Re-rating Lower）。 - [01:03:53](https://www.youtube.com/watch?v=n889nI8sR84&t=3833s) 大语言模型（LLM）等技术突破使得量化投资首次能够高效解析专利、商标和社交媒体等非结构化数据，将巴菲特式的无形资产护城河系统化量化落地。 ## Key takeaways - [12:17](https://www.youtube.com/watch?v=n889nI8sR84&t=737s) AI 基础设施建设的万亿美元级资本开支（CapEx）目前已成为全美宏观经济中唯一的确定性增长引擎，其余板块大多表现平淡。 - [21:46](https://www.youtube.com/watch?v=n889nI8sR84&t=1306s) 英伟达在绝对净利润预期已超 2200 亿美元的超大体量下，市场通过压低市盈率来消化其规模（体量折价），否则其市值将占据大盘不可思议的失衡权重。 - [26:49](https://www.youtube.com/watch?v=n889nI8sR84&t=1609s) OpenAI 与 Anthropic 正在采取类似当年 Uber 的深度补贴战策略以压低 Token 价格争抢市场份额，造成了当下的算力短缺与虚假价格信号。 - [29:47](https://www.youtube.com/watch?v=n889nI8sR84&t=1787s) 历史上的铁路、运河、电信等资本周期表明，基础设施的大规模盲目超前建设往往以产能闲置、价格暴跌和基建投资者巨亏破产告终，无伤完美交接商业 ROI 在历史上未有先例。 - [35:01](https://www.youtube.com/watch?v=n889nI8sR84&t=2101s) 传统 SaaS 软件股遭遇绝望式抛售，核心原因在于生成式 AI 击碎了代码本身的护城河壁垒，但软件巨头的真实防线在于其难以撼动的客户分发网络。 - [40:15](https://www.youtube.com/watch?v=n889nI8sR84&t=2415s) 纳斯达克 100 指数的估值结构与 1999 年互联网泡沫有着本质不同：当前 60% 的成分股市盈率处于 20 至 40 倍合理区间，且巨头拥有高达 25% 至 50% 以上的极高净利润率支撑。 - [45:47](https://www.youtube.com/watch?v=n889nI8sR84&t=2747s) 电网负荷极限与电力供应短缺构成了算力扩张的物理硬约束，但这种现实瓶颈反而在一定程度上充当了“减速器”，延缓了剧烈转型对就业和经济造成的通缩性冲击。 - [54:35](https://www.youtube.com/watch?v=n889nI8sR84&t=3275s) 重资产网络建设的历史赢家往往属于终端生态层，正如当年电信巨头砸数千亿重金建设通信网络却回报平平，最终大部分商业利润却沉淀于苹果及其生态之中。 - [57:37](https://www.youtube.com/watch?v=n889nI8sR84&t=3457s) 底层 LLM 面临快速大宗商品化的风险，头部模型间的代差被迅速抹平，行业竞争逐渐从技术绝对壁垒转向品牌心智认知与渠道分发能力。 - [01:00:42](https://www.youtube.com/watch?v=n889nI8sR84&t=3642s) 传统以市净率（P/B）为核心的价值因子在 1990 年代后逐渐失效，因其无法有效反映现代企业中以技术研发、品牌与人力资本为主导的无形资产价值。 - [01:12:49](https://www.youtube.com/watch?v=n889nI8sR84&t=4369s) 现代金融危机可能正在摆脱历时两三年的阴跌模式，转而在算法高频博弈与瞬时信息传播下演变为一到两个月内闪崩 25% 左右的极速回撤形态。 - [01:14:32](https://www.youtube.com/watch?v=n889nI8sR84&t=4472s) 从 1970 年至 2024 年长达 60 年的跨度来看，日本股市年化回报率依然接近 9%，其后三十年的长期停滞本质上是对前二十年年化 22% 狂热泡沫的极限制均值回归。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=n889nI8sR84&t=0s) 有声书的表演艺术与“没人读书”的算法时代 节目在轻松的幕后闲聊中开场。迈克尔·巴特尼克（Michael Batnick）好奇地向乔什·布朗（Downtown Josh Brown）打听录制有声书的体验。乔什坦言自己的自传体新书根本不可能找别人代读，否则听起来会极其荒谬；但他整整录了六个星期，耗尽心力。相比之下，本·卡尔森（Ben Carlson）只用了两天就搞定了自己的新书《风险与回报》（*Risk and Reward*）的有声录制。乔什打趣称自己的录音完全是一场高能演出：“我当时就像米克·贾格尔（Mick Jagger）一样在麦克风前扯着嗓子咆哮”，录完之后整个人精疲力竭。 随后话题转向了量化投资人吴凯（Kai Wu）。看到本的新书面世，吴凯也跃跃欲试，打趣说自己想写书是因为“所有酷小孩都写过书”。乔什立刻唱起反调，尖锐地发问现在到底还有没有人真正读书。作为写过四本书的老作者，乔什透露了一个行业真相：在宣传新书参与的播客访谈中，真正通读过全书的主持人大概只有10%，他一听提问就能分辨出来。大家纷纷感叹，在算法重塑人类大脑的时代，公众甚至连长文章都读不下去，只能刷刷标题，未来可能只会看关于文章的解说短视频。 - 乔什·布朗的有声书录制耗时6周，全程全情投入且风格强烈；本·卡尔森则在两天内完成录制。 - 乔什估算，他在宣传巡演中接触的播客主持人里，仅有约10%的人真正完整读完了他的书。 - 迈克尔自称是超级有声书爱好者，认为在被推荐算法改变注意力的当下，有声书成为了很多人在通勤和入眠时吸收知识的唯一途径。 ### [02:36](https://www.youtube.com/watch?v=n889nI8sR84&t=156s) 迈克尔的牛市新衣与影视闲聊：浩克·霍根的摔角传奇 插科打诨中，大家将目光转向了迈克尔的着装。乔什调侃前一天有人问迈克尔是不是换了造型师，众人笑称迈克尔最近不仅衣品大涨，而且显得“身家极为丰厚、流动性极强”（"so liquid"），完美契合当下狂热的牛市氛围。乔什直言当前的行情走势极其饥渴躁动（"The market is so horny right now"），市场情绪被彻底点燃。 接着几人聊起了最近追的流媒体热播剧与纪录片。乔什极力推荐 Netflix 上关于摔角巨星浩克·霍根（Hulk Hogan）的四集纪录片。几人回顾了霍根作为过去75年全球知名度最高、最具辨识度的人物之一的职业起伏，以及他经历过的四五次“大红大紫到人人喊打”的周期循环。乔什绘声绘色地还原了当年的名场面：霍根受西尔维斯特·史泰龙（Sylvester Stallone）之邀客串《洛奇3》（*Rocky III*）中的“霹雳火”（Thunderlips），WWE掌门人文斯·麦克马洪（Vince McMahon）雷霆大怒，威胁霍根如果敢去洛杉矶拍戏就立刻开除他，但霍根毅然前往，铸就了流行文化经典。 - 迈克尔被同僚戏称在牛市中衣品与财富同步飙升，散发着“牛市大赢家”的潇洒气质。 - 乔什大力推荐 Netflix 的4集浩克·霍根纪录片，盛赞其叙事与质感极具电影感。 - 霍根职业生涯的关键转折点：违抗文斯·麦克马洪的开除威胁，坚持参演电影《洛奇3》并饰演“霹雳火”一角。 ### [06:34](https://www.youtube.com/watch?v=n889nI8sR84&t=394s) 育儿神曲折磨与第242期正式开场 聊完霍根，众人又盘点了近期观看的影视作品，包括马修·瑞斯（Matthew Rhys）主演的新剧、乔恩·哈姆（Jon Hamm）与安娜·肯德里克（Anna Kendrick）主演的《邻家老友》（*Your Friends & Neighbors*）等。迈克尔觉得这类展现康涅狄格州富人阶层日常的剧集，虽然生活方式距离普通人很远，但人际交往中的情感细节却颇为写实。 面对大家的影视探讨，吴凯却插不上话。在迈克尔的追问下，吴凯道出了带娃奶爸的无奈：家里有个刚满一岁的女儿，他每天在电视上只能看两部剧——《Cocomelon》（可可瓜瓜）和《芝麻街》（*Sesame Street*）。乔什随即吐槽《Cocomelon》简直是无休止循环的“洗脑神曲”（slop），吴凯笑称女儿此前曾短暂迷恋过带有优质配乐的《K-pop 恶魔猎人》（*K-pop Demon Hunters*），但热度过后又重回洗脑儿歌。随着吴凯一句“让我们聊聊无形资产吧”，现场气氛从家庭育儿瞬间无缝切回资本市场硬核探讨，《The Compound and Friends》第242期正式拉开序幕。节目开篇播出了赞助商 Betterment 针对独立投资顾问（RIA）业务分层与增长策略的推介。 - 吴凯分享居家育儿日常：一岁女儿进入学步阶段，日常娱乐被《Cocomelon》和《芝麻街》完全占据。 - 众人一致对《Cocomelon》对成年人精神的洗脑摧残产生强烈共鸣。 - 本期节目的赞助商为 Betterment Advisor Solutions，重点聚焦领先 RIA 机构如何利用客户分层（Segmentation）实现业务规模扩张。 ### [08:32](https://www.youtube.com/watch?v=n889nI8sR84&t=512s) 节目开场与嘉宾介绍：投资组合管理与无形资产先锋 节目在一段 Betterment Advisor Solutions 的赞助口播后正式开启，这是全网第一投资类播客《The Compound and Friends》的第 242 期。主持人 Downtown Josh Brown 热情开场，调侃坐在身旁、衣着讲究的联合主持人 Michael Batnick，并向观众隆重介绍了本期的两位重磅返场嘉宾。当前美股市场正如火如荼，整个科技板块热度爆表，两位嘉宾的到场恰逢其时。 第一位嘉宾是 Ritholtz Wealth Management 的机构资产管理总监 Ben Carlson。Josh 戏称 Ben 根本不需要过多介绍——他与 Michael Batnick 共同主持的播客《Animal Spirits》自 2017 年 11 月开播以来长盛不衰；其个人博客《A Wealth of Common Sense》十多年来一直是严肃投资者的必读之作。Ben 拥有特许金融分析师（CFA）资格，职业生涯长期为大学捐赠基金、慈善基金会以及超长期机构配置资产。近期他刚出版了全新力作《Risk and Reward》（风险与回报）。Josh 开玩笑说“反正也没人会读”，但随即由衷赞叹 Ben 惊人的产出效率：每周不仅要更新三篇博客，还能持续出版专著。Ben 则笑称，日常写博客其实正是为他撰写新书沉淀了大量素材。 第二位返场嘉宾是 Sparkline Capital 的创始人兼首席投资官（CIO）Kai Wu。Sparkline Capital 致力于利用前沿机器学习与海量非结构化数据挖掘市场 Alpha（超额收益）。当 Josh 提及 Kai 旗下管理的两只 ETF（ITAN 和 DTAN）时，现场展开了一段关于合规限制的诙谐讨论。金融合规监管往往极其苛刻：基金经理不能在自己的播客或博客中公然推介自己的 ETF，但如果是外部媒体采访提问，则可以在合规框架内回答。Josh 随即打趣称：“Kai 只能用一段‘即兴现代舞’来确认或否认这些产品的存在了。” - **Ben Carlson 背景**：Ritholtz 机构资管主管，播客《Animal Spirits》联合主持，代表作博客《A Wealth of Common Sense》，新书《Risk and Reward》。 - **Kai Wu 背景**：Sparkline Capital 创始人兼 CIO，深耕无形资产与机器学习量化投资，旗下管理 ITAN 与 DTAN 两只 ETF。 --- ### [12:17](https://www.youtube.com/watch?v=n889nI8sR84&t=737s) AI 交易强势卷土重来与经济的“单引擎驱动” Josh 观察到，在经历了一段短暂的疑虑期后，AI 主题交易迅速重新占据华尔街舞台的绝对中心，成为了“全市场唯一的焦点”。Josh 直言，在全美经济体系中，除了由 AI 基础设施建设引发的万亿美元级资本开支（CapEx）热潮之外，几乎看不到其他明显的增长亮点。整个宏观经济的其他板块要么平淡无奇，要么表现低迷，而任何能够实现强劲增长的细分行业，追根溯源都与这场万亿规模的 AI 算力军备竞赛紧密挂钩。 然而，在重回巅峰之前，AI 叙事其实曾遭遇过严重危机。此前市场情绪极度悲观，甚至出现了像甲骨文（Oracle）这样的大溃败。起因是 OpenAI 首席执行官 Sam Altman 在 Brad Gerstner 的播客节目中发表了一系列引发恐慌的言论，市场随即质疑这些初创 AI 公司究竟能否兑现向甲骨文承诺的为期 5 年、高达 3000 亿美元的云服务巨额采购合同，导致甲骨文股价一度自高点暴跌近 60%。但随后，大语言模型（LLM）的性能以惊人的速度迭代进化，Anthropic 等领头羊企业不断公布突破天际的营收数字，甲骨文的股价也应声反弹，创下数月以来的新高。市场终于意识到，底层 AI 公司的商业化收入正在以真实且狂暴的节奏兑现，先前的合同风险正在被超预期的基本面彻底化解。 ```mermaid flowchart LR accTitle: AI交易逻辑反转流程 accDescr: 展示市场从恐慌抛售到因基本面大增而重新狂热的演变逻辑 subgraph Fear ["恐慌阶段"] A["高管言论引发质疑<br/>（初创公司能否兑现承诺）"] --> B["算力客户违约担忧<br/>Oracle 股价大跌 60%"] end subgraph Turnaround ["转折与爆发"] C["模型能力突飞猛进<br/>Claude 3.5 Sonnet / Opus"] --> D["商业化变现超预期<br/>季度收入指数级飙升"] end subgraph Mania ["全面狂欢"] D --> E["甲骨文股价强劲反弹<br/>重回多月高位"] E --> F["AI CapEx 成为全美<br/>唯一核心经济增长引擎"] end Fear --> Turnaround ``` --- ### [14:26](https://www.youtube.com/watch?v=n889nI8sR84&t=866s) Anthropic 疯狂的基本面：季度营收翻三倍与 500% 净留存 Kai Wu 与 Michael Batnick 深入拆解了 Anthropic 旗下 Claude 模型在开发者群体中所掀起的风暴。Kai 指出，ChatGPT 在 2022 年底点燃了面向普通消费者的第一波浪潮，而 Anthropic 随后推出的 Opus 以及 Claude 3.5 Sonnet 与 Claude Code 工具，则精准引爆了全球软件工程师圈子。这直接改写了整个产业叙事，推动 Anthropic 的商业化数据呈现出让所有人“下巴脱臼”般的裂变式扩张。 Michael 引述了对冲基金投资人 Patrick O'Shaughnessy 对 Anthropic 首席财务官（CFO）Krishna Rao 专访中所披露的一系列令人瞠目结舌的真实运营指标： - **净美元留存率（NDR）年化超 500%**：通常软件企业 100% 的留存率代表客户没有流失，而 500% 意味着现有老客户在一年内追加消费了 5 倍的金额。 - **商业化起点极短**：Anthropic 产生第一笔商业收入的时间居然是在 2023 年 3 月。 - **内部代码自动化**：Anthropic 内部超过 90% 的软件代码均由 Claude Code 自主编写生成。 - **重度算力消耗场景**：在财务团队中，消耗 Token 数量最多的核心使用者竟然是税务总监（Head of Tax）。 - **营收运行率（ARR）狂飙**：公司季度营收运行率在一个季度内从 90 亿美元垂直拉升至 300 亿美元，环比实现数倍增长，并有望在次月冲上 500 亿美元。 - **新产品增长极值**：内部新推出的协同工具 Cowork，其早期的增长曲线甚至超越了同期爆火的 Claude Code。 - **签约速度前所未见**：CFO 透露，在坐 Uber 前往录制该期播客的短短 20 分钟车程内，他就通过手机线上签署了两笔千万美元级（Double-digit million dollar）的巨额企业级采购订单。 面对这样史无前例的指数级增长，主持人们惊叹之余也保持了理性。Ben 与 Josh 指出，如果按照这个速度继续复合年化两年，Anthropic 的体量将超越美联储的资产负债表并成为全球最大公司，因此增速未来必然会放缓。但不可否认的是，整个商业生态正在经历剧烈洗牌——这数以百亿计的现金流并非凭空凭空印出来的，而是企业软件开发、对冲基金量化投入、大型金融机构（如贝莱德 BlackRock）以及工业跨国巨头从其他旧 IT 预算中大规模转移出来的重金再分配。 --- ### [16:26](https://www.youtube.com/watch?v=n889nI8sR84&t=986s) Cerebras IPO 盛宴：百倍市销率与算力抢购狂潮 AI 热潮在资本市场的极致投射不仅体现在一级市场未上市的 Anthropic 上，更直接点燃了公开股票市场的首次公开募股（IPO）。Josh 分享道，就在当天早些时候他造访纽约证券交易所（NYSE）时，贝莱德与 Digital Realty Trust 的相关大型基建交易正在备受瞩目，而纳斯达克则迎来了另一家芯片与算力巨头——Cerebras Systems 的震撼挂牌。 现场嘉宾对 Cerebras 的 IPO 估值与市场抢筹情绪进行了讨论。根据场内初步反馈的数据： - **募资规模**：Cerebras 本次上市募资约 55 亿美元，公开发行了 3,000 万股股票，并附带 450 万股绿鞋超额配售选择权（Greenshoe Option）。 - **二级市场表现**：发行定价约为 185 美元，但开盘价直接飙涨至 350 美元，随后在 326 美元附近高位震荡。 - **估值水平**：对应其开盘后股价，公司市值迅速推升至接近 1,000 亿美元级别；而其年化经常性收入（ARR）大约刚达到 10 亿美元水平，相当于以超过或接近百倍的市销率（P/S）进行估值交易。 - **火爆认购比例**：其机构认购订单薄（Order book）获得了高达 20 倍的超额认购。在开盘集合竞价撮合时，市场上流通的每一股卖盘对应着多达 5 个争先恐后的买入订单。 尽管嘉宾们坦言对该公司的底层微观业务细节仍待深入观察，但如此疯狂的资金追捧再次印证：只要沾上 AI 核心底层算力的概念，华尔街的买方资本就会不惜一切代价涌入，推动估值彻底脱离传统的财务模型常识。 ### [17:52](https://www.youtube.com/watch?v=n889nI8sR84&t=1072s) Cerebras 的 IPO 盛宴与推理芯片的竞争挑战 讨论聚焦于刚刚推向公开市场的芯片公司 Cerebras。主持人提到，散户投资者正在疯狂涌入，但实际上所有大型机构早已提前布局。如今的风投生态中，企业在真正公开上市之前就已经经历过多轮“准 IPO”融资，甚至进行到了 H 轮；像富达投资（Fidelity）这样的大型机构在 IPO 前就已经重仓买入。这标志着自 2020 年 9 月 Snowflake 募资 38 亿美元以来美国规模最大的科技股 IPO——尽管 Snowflake 当时以超过 100 倍市销率（P/S）的高估值上市，随后股价经历了剧烈暴跌。 Cerebras 最具看点的故事在于其与 OpenAI 签订的主协议：OpenAI 锁定了 750 兆瓦（MW）的推理算力，并计划到 2030 年扩展至 2 吉瓦（GW）。实际上，该公司一年半前曾提交过 S-1 招股书，但随后撤回，主要原因是市场对极高的客户集中度感到担忧——当时投资者对一家营收几乎全系于两三家巨头（包括 OpenAI 和阿联酋的大单）的半导体公司心存疑虑。然而时过境迁，在当下的市场情绪中，客户集中度已不再被视为不可逾越的风险。 在技术路线上，Cerebras 并非制造微小的纳米芯片，而是打造整块巨型晶圆级别的芯片（Wafer-Scale Engine）。其核心设计理念在于将内存紧密贴近算力单元，以此消除内存与计算之间的传输延迟，号称是用于推理（inferencing）的终极架构。Kai Wu 分析指出，这完全符合产业资本周期的演进逻辑：英伟达（Nvidia）享有极高的营业利润率，并且在庞大体量下仍保持爆炸式增长，这种丰厚的利润空间自然会吸引各方入局竞争，无论是初创企业还是谷歌、亚马逊等正在自研芯片的云厂商（Hyperscalers）。 ```mermaid flowchart TD accTitle: Cerebras 业务与资本周期逻辑 accDescr: 展示 Cerebras 芯片架构优势、关键大客户协议以及推动行业竞争的资本动力。 subgraph TechArch["核心技术特点"] W1["整块巨型晶圆 (Wafer-Scale)"] W2["内存紧邻计算核心"] W3["极致消除推理延迟"] W1 --> W2 --> W3 end subgraph Customers["关键大客户"] C1["OpenAI 750MW 推理协议<br/>(2030年可扩展至 2GW)"] C2["阿联酋大额商业协议"] end subgraph MarketDynamics["资本周期动力"] M1["英伟达超高利润率与爆发式增长"] M2["新入局者争相进入<br/>(Cerebras、谷歌、亚马逊等)"] M1 --> M2 end TechArch --> Customers Customers --> MarketDynamics ``` ### [20:15](https://www.youtube.com/watch?v=n889nI8sR84&t=1215s) 估值锚点：27 倍市盈率与 7 天 9000 亿美元的暴涨 分析师常常被质疑在牛市中沦为盲目看多的“啦啦队长”（cheerleader）。有人向现场嘉宾提问，当下的 AI 热潮与当年的互联网泡沫（Dot-com bubble）到底有何区别？嘉宾回应道，最大的区别在于当年的财经媒体更加狂热，而如今的行情完全有坚实的盈利增长作为支撑。以半导体行业 ETF（SMH）为例，虽然自低点以来累计涨幅超过 220%，但其远期市盈率（Forward P/E）仅为 27 倍左右。尽管半导体绝不算便宜，但 27 倍的估值倍数与 1999 年动辄上百倍估值的狂热泡沫有着本质区别。 然而，英伟达的市值扩张速度已经逼近现实认知的边界。在经历此前漫长的横盘整理后，英伟达在仅仅 7 个交易日内市值就飙升了 9000 亿美元。现场嘉宾幽默地列举道，这 7 天内凭空增加的市值，相当于麦当劳（McDonald's）、迪士尼（Disney）、波音（Boeing）、优步（Uber）、星巴克（Starbucks）以及皇家加勒比邮轮（Royal Caribbean）六大蓝筹企业的市值总和，“这些数字庞大到几乎快失去现实意义了”。 - **横盘消化期**：在这一波爆发前，英伟达股价曾有长达数月在 180 美元至 200 美元区间震荡，甚至此前在 235 美元附近停滞近一年，给市场留足了“消化”业绩的时间。 - **超预期财报反杀**：此前在奥斯汀录制节目期间，英伟达公布了一份极其亮眼、堪称夸张的财报，结果次日股价开盘上涨 4% 随后收盘倒跌 4%，令不少投资者误以为“这一轮交易已经结束”。 ### [21:46](https://www.youtube.com/watch?v=n889nI8sR84&t=1306s) 体量折价：英伟达无法享有超高溢价的数学逻辑 关于英伟达应当享受多高的估值，嘉宾们从指数权重的数学角度给出了深刻剖析。如果一家公司的净利润只有 200 亿美元，给予其 50 到 60 倍的远期市盈率在数学上仍能被市场容纳；但根据预测，英伟达未来 12 个月的净利润预期已经高达惊人的 2230 亿美元。当绝对利润规模庞大到这种地步时，它在逻辑上就失去了享受过高估值溢价的物理空间。 数据图表显示，在 25 倍远期市盈率下，英伟达的市值约为 6 万亿美元左右，已占到标普 500 指数权重的 8.5%。如果市场给予其 45 倍的远期市盈率，英伟达的市值将膨胀至 10 万亿美元，单一股票就将独占整个大盘指数 15% 的权重。届时整只指数几乎沦为“英伟达与 Anthropic 两家公司的游戏”。因此，英伟达必须承担所谓的“体量折价”（Size discount），市场在业绩不断刷新新高时通过压低市盈率来消化其规模，是完全合理的制度自我约束。 ### [24:18](https://www.youtube.com/watch?v=n889nI8sR84&t=1458s) 单一交易链条与企业 AI 军备竞赛 有观点甚至认为英伟达已经不再是一只单一股票，而是蜕变成了一个独立的“板块”（accelerated computing）。针对在投资组合中将其配置为 10% 权重的激进做法，Kai Wu 指出必须清醒认识到半导体行业的固有周期性与竞争格局。不仅 Cerebras 这样的新贵正在发起进攻，英伟达最大的客户群（谷歌和亚马逊等超大规模云服务商）也在不遗余力地自研芯片。历史上，英伟达长期是波动剧烈的强周期股，如今的高增长同样锚定在一个宏大的宏观闭环中：云厂商百亿美元级的资本支出（CapEx）投向数据中心建设，进而滋养了芯片制造商和电力公用事业公司——本质上，这全都是“同一笔交易”。 这引发了一个根本性追问：下游企业究竟能否顺利完成 AI 落地，并将其转化为真金白银的业务增长？当前 Anthropic 飞速攀升的营收年化运行率（run rate），很大程度上源于董事会对管理层的问责压力——董事会明确警告 CEO 们“如果不拥抱 AI 进行数字化转型，那你就会被解雇”。整个美国企业界因此陷入了一场唯恐落后的军备竞赛。如果企业应用后能带来显著的营收提振，这笔投资将持续存在；若证明只是收效平平的试错，资本支出终将面临削减。 Josh Brown 分享了发生在自己公司的鲜活趣事：前一天，负责为其公司搭建数据湖（Data Lake）的技术公司 Invent 来访，管理层齐聚会议室开会。Josh 推门而入直奔主题：“各位，别绕弯子了，我们到底有没有搞 AI？”（"Guys, let's cut to the chase. Do we do an AI?"）对方答复道：“是的，我们正在搞 AI。”Josh 随即满意离开并表示：“很好，太棒了。”这虽然是一句玩笑，却真实折射出当前整个商业世界的常态：每一个行业、每一家公司的领导层，都在急迫地向外界展示自己“正在布局 AI”，以此证明公司依然站在时代的前沿。 ### [26:49](https://www.youtube.com/watch?v=n889nI8sR84&t=1609s) 企业端 FOMO 与模型公司的“Uber 式补贴战” 企业管理层目前正面临前所未有的汇报压力：无论是面对董事会还是股东，高管们必须明确汇报“因为引入了这项技术，我们多赚了 X 美元，或者削减了 X 美元成本”。在这种紧迫感下，AI 相关的资本支出完全不再是可选项目，而是被外部环境倒逼的必然选择。 对于市场上普遍讨论的“AI 算力瓶颈”以及 Anthropic 算力受限的问题，Kai Wu 指出，这种现象背后叠加了双重力量：一是企业层面的错失恐惧症（FOMO），二是 Anthropic 与 OpenAI 之间你死我活的份额争夺战。这两家模型公司正在沿用当年 Uber 的经典打法——坚信市场属于“赢家通吃”（Winner-takes-all）或“赢家通吃大部分”，为了抢占先机，不惜通过深度补贴人为压低 Token 价格。在亏损状态下激进抢占市场份额，意味着当下的算力短缺并不能反映真实的供需价格信号。核心问题应当转向：AI 技术究竟能否为客户创造切实的投资回报率（ROI）？如果能，其中的商业价值最终会被模型实验室、超大规模云服务商（Hyperscalers）拿走，还是会真正沉淀给下游终端客户？ ```mermaid flowchart TD accTitle: AI生态价值竞争图 accDescr: 展示了从企业FOMO与价格补贴到最终商业回报的核心链条 subgraph Inputs["驱动力量"] A["企业高层 FOMO<br>必须向董事会证明 ROI"] B["模型厂商市场战<br>OpenAI vs. Anthropic"] end subgraph Dynamics["市场扭曲机制"] C["Uber 式打法<br>亏损抢占市场份额"] D["补贴压低 Token 价格<br>产生虚假价格信号与算力挤兑"] end subgraph Question["最终核心拷问"] E{"AI 能否真正驱动企业 ROI？"} F["价值分配归属：<br>超大规模云厂商 vs. 模型实验室 vs. 终端客户"] end A --> C B --> C C --> D D --> E E --> F ``` --- ### [28:09](https://www.youtube.com/watch?v=n889nI8sR84&t=1689s) 量化分析师眼中的 AI 落地与任务进阶 谈及自身体验，Kai Wu 认为 AI 的实用性毋庸置疑。作为早在 2019 至 2020 年就开始训练大语言模型（LLM）的早期采用者，他坚信这项技术是货真价实的变革，尽管目前依然不能在所有关键环节盲目信任它。 商业颠覆的历史规律表明，新技术往往从容错率极高的低端任务切入。在量化投资领域，Kai 过去通常指派初级分析师去跑因子测试模型或在不同市场中做回测；如今，像 Claude Code 或 OpenAI Codex 这样的工具已经完全可以接管这类初阶分析工作。只需给模型提供 API Key、数据库架构以及示例脚本，它就能迅速模仿并输出分析结果。现场嘉宾调侃 Kai“穿了西装夹克也掩盖不住量化男的本质，梳着丸子头的量化极客更是罕见”。Josh Brown 与 Michael Batnick 观察到，人机交互的模式已经从“量化主管给下属分配任务”彻底转变为“直接与模型对话沟通”的即时问答迭代过程。 --- ### [29:47](https://www.youtube.com/watch?v=n889nI8sR84&t=1787s) 历史资本周期与“这次不一样”的认知失调 Michael Batnick 提出了一个困扰他数月的认知失调：Kai Wu 曾在 GMO 与传奇投资人格兰瑟姆（Jeremy Grantham）以及历史学家爱德华·钱瑟勒（Edward Chancellor，《资本的回报》《抢滩投机史/Devil Take the Hindmost》作者）共事。Chancellor 在资本周期理论中明确指出，每一轮重大的资本支出狂潮（CapEx Boom）都会在过度开支后以惨烈暴跌收场；而约翰·邓普顿（John Templeton）又曾说过，大约有 20% 的情况下“这次真的不一样”。 针对这种历史规律与当下乐观情绪的冲突，Kai 表示两者完全可以同时成立：**AI 技术本身极具变革性，但它在当前周期里依然可能是一笔糟糕的投资。** Chancellor 研究过电力、运河、铁路以及 2000 年互联网泡沫等历次基建狂潮，除了当年因自然垄断而整合成功的电信电话行业之外，历史上几乎每一次都是资本盲目涌入供给侧、疯狂铺设基础设施，最终因需求释放跟不上供给节奏而出现“空气断层”（Air Pocket）。价格崩溃与大量闲置产能接踵而至，具有讽刺意味的是，那些最早研发出核心技术并大举建设基础设施的企业大多在随后的洗牌中走向破产。 - 资本周期的经典铁律：大量资本盲目流入供给侧，基础设施超前建设，随后需求跟进滞后导致价格暴跌与产能闲置。 - 历史惨烈先例：当年成百上千家铁路公司、初创汽车公司和电信宽带铺设商，都在建设潮退去后倒闭破产。 --- ### [32:08](https://www.youtube.com/watch?v=n889nI8sR84&t=1928s) 囚徒困境、贝佐斯的“工业泡沫论”与完美交接之辩 当 Michael 质疑现代企业管理者拥有比百年前强得多的数据洞察力、不至于突然集体陷入“哎呀搞砸了”的盲目局面时，Kai 驳斥指出，这是典型的行为金融学与博弈论问题。企业管理层身处经典的“囚徒困境”（Prisoner's Dilemma）：扎克伯格（Mark Zuckerberg）等科技巨头 CEO 甚至公开表态“宁可过度投资破产，也绝不能在这场竞赛中落后”。对于巨头而言，一旦 Sam Altman 带领 OpenAI 率先做出了通用人工智能（AGI），坐以待毙的企业就直接面临生死存亡。因此，Sam Altman 成了推高军备竞赛的领头羊，所有人都被绑架着跟注。 Michael 引用杰夫·贝佐斯（Jeff Bezos）的评论指出，贝佐斯也认为 AI 虽对社会深远有利，但本质上是一场“工业泡沫”，当前巨额投资中的大部分最终将被白白浪费。面对 Michael 提出的二选一考验——“究竟是像互联网泡沫那样经历破灭但技术最终大获成功更令人吃惊，还是巨额资本开支无缝交接到实际商业 ROI 的‘完美接棒’更令人吃惊？”——Kai 毫不犹豫地表示后者几乎不可能发生。在人类商业史上，如此庞大体量的资本狂潮从未有过“完美无伤交接”的先例；基础设施投资者所面临的下行风险敞口极其不对称。 --- ### [34:10](https://www.youtube.com/watch?v=n889nI8sR84&t=2050s) 市场狂欢下的极度分化：1929、1973 与 1999 的阴影 讨论随后切入具体市场指标。Josh Brown 提到了市场分析师 Jason Goepfert 发布的一项惊人数据：标普 500 指数在创下历史新高的同时，竟然有超过 5% 的成分股跌至 52 周新低。在美股历史上，这种极端分化的现象此前仅发生过三次，分别对应着标志性的周期顶部： | 历史极端背离时间 | 后续市场走向背景 | |:--- |:--- | | **1929 年 7 月** | 大萧条前夕的美股大崩盘起点 | | **1973 年 1 月** | 伴随滞胀与中东石油危机的漫长熊市 | | **1999 年 12 月** | 互联网泡沫见顶与崩盘前夕 | 在道琼斯指数向 50,000 点发起冲击的狂欢派对幕后，大量股票正在被无情抛弃。创下 52 周新低的阵营高度集中在房地产产业链、建材家居巨头（如 Home Depot）、餐饮以及大量基础软件（SaaS）公司；与此同时，标普 500 等权重指数（Equal Weight）涨幅明显滞后（仅上涨约 7%）。这种少数科技权重大涨托起指数、实体经济受高利率压制、底层消费与软件股大面积失血的状态，正是极具危险意味的“K 型经济”与市场广度严重恶化的真实写照。 ### [035:01] 软件股崩盘与“代码护城河”的幻灭 在谈到当前科技股上涨广度不足、呈现出强化版“K型经济”（K-shaped economy on steroids）时，嘉宾们探讨了半导体与软件股的剧烈分化。目前半导体板块持续暴涨，而软件股却在遭遇无情抛售，两者几乎变成了镜像交易。市场普遍存在一种叙事逻辑：人工智能越强大，对传统软件行业的颠覆就越彻底。Josh Brown 直言不讳地指出，软件公司的走势图令人绝望，他甚至感觉大量软件公司正在走向归零（going to zero），就像当年的传统纸媒一样彻底丧失未来。虽然并非每一家公司都会真的破产清算，但很多上市软件公司的股价可能永远无法重回巅峰。他翻看了大约 20 家软件公司的周线图，几乎全线逼近或处于 52 周新低，不少股票甚至出现了连续 25 周周五收盘价低于周一开盘价的罕见跌势，且市场上几乎没有任何并购私有化的动静。 ```mermaid flowchart LR accTitle: AI对软件与半导体板块的影响逻辑 accDescr: 展示生成式AI如何通过降低代码成本颠覆传统软件公司，同时推动半导体板块市值扩张的过程。 subgraph Shift["资本与叙事分化"] AI["生成式 AI 崛起<br/>(编写代码边际成本趋零)"] SOX["半导体板块 (SOX)<br/>占标普500权重达23%"] Soft["传统SaaS与软件股<br/>估值重估，逼近52周新低"] end AI -->|"硬件需求爆发<br/>算力成为基础设施"| SOX AI -->|"打破代码护城河<br/>分发渠道受咨询大军冲击"| Soft ``` 针对 Adobe 这类代表性软件公司能否幸存的疑问，Kai Wu 分析了市场暴跌的核心原因：投资者过去一直认为软件公司的核心护城河是“代码”。然而在当今由云端 AI 驱动的代码生成环境下，写代码的成本几乎降为零，任何非专业开发者都有能力通过自然语言生成应用来替代 Salesforce 或 Adobe 的某些功能。Kai Wu 进一步反思，代码或许从来都不是这些公司的真正护城河，长久以来的核心壁垒其实是客户分发网络（distribution）。面对 OpenAI 和 Anthropic 等巨头是否能用资本颠覆这些分发渠道的追问，Kai 指出 AI 初创公司已经意识到了企业软件不能仅靠“只要产品酷炫，企业客户就会不请自来”的极客思维。OpenAI 正在联合大量私募股权机构建立部署前线工程师团队，扮演咨询顾问的角色深入企业推销方案，并与埃森哲（Accenture）等传统 IT 咨询机构建立深度合作，同时 Google 也在大举招募类似的销售与服务团队，直接攻克企业决策层。 - 软件股技术面遭遇灾难性抛售，多只标的呈现连续 25 周下跌、缺乏任何实质性反弹行情的极端走势。 - 市场悲观情绪认为软件公司缺乏技术壁垒，但讨论指出真正的壁垒可能在于深固的企业分发体系与惯性转换成本。 - AI 头部企业从早期被动的“筑巢引凤”心态，转型为雇佣大量现场顾问、联合 PE 和咨询巨头强推企业级落地的主动直销策略。 --- ### [038:03] 估值倒挂与半导体交接棒：消失的软件市值 Michael Batnick 与 Josh Brown 随后对比了当前软件公司的实际业绩与极端估值之间的脱节现象。令人费解的是，诸如 Adobe 等头部软件巨头的每股收益（EPS）和每股自由现金流依然维持在历史最高水平，Salesforce 的底层业务也并未发生实质性的恶化，但市场却早已对当下的现金流视若无睹，股价动辄从高点下跌 60%。Adobe 的市盈率（P/E）一度被杀到 9.5 倍甚至 10 倍左右，这种估值倍数通常只属于重资产、周期性极强的传统传统汽车制造商（Auto maker），而不是高利润率的软件企业。出现这种倒挂的唯一合理解释是：市场认定这些软件公司明年的利润将会断崖式崩塌，当下的盈利只是假象。 Ben Carlson 将这种现象比作美国整体经济中的“滚动式衰退”（rolling recession）——个别行业和输家正在惨烈出清，但并未拖垮整个大盘。半导体行业膨胀的市值远远超过了软件行业蒸发的市值，这对纳斯达克和标普 500 指数而言仍然是净利好。费城半导体指数（SOX）涵盖的芯片公司在标普 500 中的权重已飙升至 23%，仿佛完成了一场巨头之间市值的“交接棒”。尽管私募股权与私人信贷领域手握超过 3 万亿美元的“干火药”（dry powder），但由于 PE 机构自身此前早已重仓软件资产、风险敞口极高，面对以极低估值交易的上市公司，目前尚未出现大规模的杠杆收购（LBO）或私有化动作。 - 软件龙头估值极具冲击力：拥有强劲现金流的高毛利软件企业估值压缩到 10 倍左右 P/E，相当于传统车企的估值水平。 - 标普 500 指数呈现剧烈结构重组：费城半导体指数权重飙升至 23%，25 家软件巨头的萎缩被另外 25 家半导体与硬件新贵的大幅膨胀所覆盖。 - 资本市场面临估值未解之谜：PE 机构手握 3 万亿美元现金储备却按兵不动，因其在软件投资组合中的敞口早已处于过度饱和状态。 --- ### [040:15] 纳斯达克盈利神话与二十年 14% 的复利奇迹 对话随后转入纳斯达克市场的基本面支撑。许多人习惯将当下的科技狂热与 1999 年的互联网泡沫进行对比，但 Michael Batnick 与 Ben Carlson 通过数据指出，当下的科技繁荣其实是“史上最合乎逻辑的一次暴涨（the most logical melt-up of all time）”。根据纳斯达克 100 指数的历史财报统计：在 1999 年科网泡沫顶峰时期，指数中高达 10.5% 的成分股净利润为负，没有任何一家公司的利润率能够达到 50% 至 100% 的区间，大多数企业利润率徘徊在 0% 到 25% 之间；而如今，纳斯达克 100 指数中有 20% 的公司利润率处于 50% 到 100% 的惊人水平，整整 50% 的成分股利润率在 25% 到 50% 之间。这证明今天的巨头是拥有庞大现金流与垄断级盈利能力的优质资产，与当年靠炒作无营收概念股的投机狂欢有着本质区别。 | 指标维度 | 1999 年科网泡沫顶峰 | 当期（纳斯达克 100 指数） | |:--- |:--- |:--- | | **净利润为负的成分股比例** | 10.5% | 极低（优质成熟企业为主） | | **利润率在 50% - 100% 的成分股比例** | 0% | 20% | | **利润率在 25% - 50% 的成分股比例** | 24% | 50% | | **利润率在 0% - 25% 的主流成分股** | 绝大多数公司集中于此区间 | 显著收缩，少数成熟低利润企业 | | **企业基本面特征** | 存在大量预营收（pre-revenue）与投机性 IPO | 全球化自由现金流巨兽，具备极高资本回报率 | 虽然 Dan Greenhaus 曾提醒大家不要神话当下而过分贬低 1999 年——当时思科（Cisco）营收达 190 亿美元，微软（Microsoft）营收达 220 亿美元（按通胀调整后分别相当于现在的 380 亿美元和 440 亿美元，堪比当下的巨头体量）——但当时指数充斥着大量毫无收入的空壳公司，而目前完全缺乏这种劣质 IPO 潮。更具冲击力的数据是纳斯达克 100 指数在过去 20 年间录得了年化 14% 的净利润复合增长率，支撑了指数年化近 20% 的总体涨幅。面对未来利润能否持续高增长的疑问，嘉宾们探讨了驱动下一个十年的核心引擎：除了生成式 AI，人形机器人与全球无生命物理对象的全自动化改造，乃至生物科技的黄金时代（如寿命延长至 120 岁），都可能成为支撑下一阶段 15% 营收增长的强力动能。Josh 总结道，在剧烈的自动化转型与技术跃迁浪潮中，年轻投资者最糟糕的策略莫过于“年纪轻轻却持有过多现金”；一旦遭遇非系统性衰退的市场暴跌，场外资金在短短几秒钟内便会迅速涌入逢低扫货，再次形成强劲的“V 型反转”，因为没有任何资本承担得起错过这轮自动化革命的代价。 ### [44:01](https://www.youtube.com/watch?v=n889nI8sR84&t=2641s) 纳斯达克估值对比：1999 年泡沫与当下的本质区别 在讨论市场前景时，嘉宾们指出，尽管市场肯定会出现回调，但在当前如此强大的技术创新浪潮下，很难想象会出现长达两年的漫长熊市。回顾过去几个月的个股波动，美光科技（Micron）就是一个典型的极端例子：3 月公布财报后，其股价短短 10 个交易日内暴跌 30%，从 470 美元跌至 310 美元，但随后又在短短一两个月内一路飙升至 800 美元。这种剧烈反转说明，一旦当前的周期逻辑发生转向，市场变脸的速度会极快。 主持人调出对比图表，回顾 20 世纪 90 年代纳斯达克 100 指数的走势。在当年的超级狂热（Melt-up）之后，纳斯达克指数最终暴跌 83%，耗费整整 12 年才重回盈亏平衡点。然而，大家一致认为历史不会简单重演，今天几乎不可能再出现这种幅度和时长的崩盘，核心原因在于估值结构的根本差异： - **1999 年互联网泡沫时期**：指数中 32% 的成分股市盈率（P/E）在 60 至 100 倍之间，高达 34% 的成分股市盈率超过 100 倍，且有 10% 的公司完全没有盈利。 - **当前市场格局**：60% 的成分股市盈率处于 20 至 40 倍的合理区间，20% 的成分股在 40 至 60 倍之间，只有 10% 左右的成分股估值处于脱离基本面的荒谬水平。 当前的估值水平与 1999 年有着本质不同。市场腰斩（回调 50%）随时可能发生，但要再次暴跌 80% 则极难令人信服。 --- ### [45:47](https://www.youtube.com/watch?v=n889nI8sR84&t=2747s) 市场核心风险：集中度泡沫、就业冲击与电网瓶颈 谈到当前市场面临的最大风险，嘉宾们首先指出了极高的指数集中度。目前美股七巨头（Mag 7）占标普 500 指数权重的 33%，若算上核心芯片股，被动指数中高达 50% 至 60% 的权重实质上都捆绑在同一个主题——AI 之上。被动投资原本被视作分散风险的工具，现在却变成了对单一押注的高度集中。一旦科技巨头削减资本开支，科技股与实体经济深度绑定将可能同时引爆经济衰退和股市熊市。 ```mermaid flowchart TD accTitle: AI 浪潮下的潜在风险传导路径 accDescr: 展示集中度风险、就业冲击与电网电力瓶颈如何影响市场与经济 subgraph MacroRisk["市场与宏观风险"] A["指数高度集中 (Mag 7 + 芯片股占 50-60%)"] --> B["资本开支削减 / 叙事受挫"] B --> C["熊市与经济衰退双生"] end subgraph StructuralBottlenecks["结构性制约因素"] D["AI 效率超预期爆发"] --> E["大规模裁员与就业替代"] F["电网负荷与电力短缺"] --> G["算力扩张物理硬约束"] G -. 充当减速器 (Governor).-> D end ``` 关于 AI 带来的颠覆性风险，讨论聚焦于三种主要可能性： 1. **劳动力市场替代**：有人担心 AI 技术突破过快导致裁员潮，使就业市场承受巨大冲击。尽管有人开玩笑说应对这种焦虑的方法就是“买股票，别被机器人打败”，但 Kai Wu 认为这种风险在短期内被过度放大了，大众普遍受科幻电影影响高估了技术落地的即时冲击。 2. **电力与电网瓶颈**：随着算力需求膨胀，制约资本开支狂欢的不仅是芯片短缺，还有电网的承载能力。 3. **“减速器”（Governor）效应**：Kai Wu 提出，物理瓶颈的存在或许是一件好事。如果 AI 变革在一夜之间完成，政府监管、员工再培训和企业重组根本无法跟上，极易陷入恶性通缩循环；而电力短缺与供应链瓶颈就像一台“减速器”，让资本建设在相对长的时间里逐步释放生产力，从而实现平稳过渡。 --- ### [48:51](https://www.youtube.com/watch?v=n889nI8sR84&t=2931s) 从轻资产到重资产：Mag 7 的估值重估陷阱 Kai Wu 提出了一个核心基准假设：七巨头不会倒闭破产，但面临着估值中枢下调（Re-rating Lower）的根本性风险。过去 15 到 20 年间，七巨头之所以能够享有远超常态的估值溢价，是因为它们凭借品牌、人才、网络效应等“无形资产”（Intangible Assets），在万亿美元级的体量上实现了史无前例的高资本回报率（ROIC）。 Alphabet（Google）就是典型代表：25 年前发明了搜索引擎，凭借近乎不需要追加重度资本的轻资产模式享受了数十年的超高利润率，随后的云计算业务同样具备极高毛利。然而，当下的 AI 军备竞赛彻底改写了这一游戏规则： - **博弈论困境（囚徒困境）**：在 AI 时代之前，巨头们各自割据（搜索归 Google、社交归 Meta、电商归 Amazon），维持着舒服的寡头格局。但随着 AI 智能体（Agents）的出现，市场格局可能面临“赢家通吃全部入口”的重塑。为了防范这一生存危机，哪怕萨姆·奥特曼（Sam Altman）等人提出投入万亿美元，其他巨头也只能被迫跟进，陷入了停不下来的资本竞赛。 - **由轻转重的商业模式转变**：为了支撑海量计算，科技巨头正在疯狂投资于数据中心、电力设施和物理硬件，从当年轻盈的无形资产公司蜕变为重资产运营方。 - **公共事业化与折旧压力**：当市场逐渐意识到这些科技霸主不再是 5 年前的高利润轻资产软件企业，而变成了背负庞大硬件折旧负担的“高科技公共事业企业（Utilities）”时，维持当前创历史新高的估值倍数将变得极为困难。大规模资本支出转化为资产负债表上的固定资产后，随之而来的巨额折旧将直接侵蚀未来的盈利表现。 ### [52:25](https://www.youtube.com/watch?v=n889nI8sR84&t=3145s) 资本支出的会计戏法与电信巨头的“惨胜”教训 对话从数据中心资本支出的会计处理切入。按现行会计准则，购买数据中心所花费的资本支出（CapEx）被计入资产负债表，并在未来几年内逐步折旧。以 5 年期折旧计算，科技巨头每花 1 美元建设数据中心，当年计入净利润的折旧成本仅约 20 美分，然而英伟达（Nvidia）却能立即将这 1 美元全额确认为当期营收。这种时间差造成了买方净利润在会计层面被人为平摊修饰，而卖方则瞬间兑现暴利。 ```mermaid flowchart LR accTitle: 资本支出流向与利润确认时间差 accDescr: 超大规模科技公司资本支出通过折旧缓慢计入成本，而英伟达全额即时确认营收 subgraph Buyer ["超大规模云厂商 (Hyperscalers)"] direction TB B1["1 美元 CapEx 支出"] --> B2["资产负债表资本化"] B2 --> B3["5 年折旧: 每年仅 20 美分入利润表"] end subgraph Seller ["硬件提供商 (Nvidia)"] direction TB S1["全额 1 美元营收"] --> S2["当期净利润暴增"] end B1 -.->|"采购 GPU / 算力"| S1 ``` 乔什·布朗（Downtown Josh Brown）随即指出，庞大的资本支出周期对股东而言往往是极其残酷的。他以美国电信业作为典型反例：从 1G 一路建设到 5G，电信公司耗费了长达 25 年的时间以及成百上千亿美元的无底洞投资，然而 AT&T 和 Verizon 的股价如今竟然与 1997 年毫无二致。虽然迈克尔·巴特尼克（Michael Batnick）打趣说“至少拿着股息等待了”，但乔什反驳称，当年还叫大西洋贝尔（Bell Atlantic）的 Verizon 在 1998 年是 45 美元，现在的股价依旧是 45 美元，就算有 5% 的股息率，它充其量也只是一张伪装成股票的债券。整个美国移动通讯行业最终只剩 AT&T、Verizon 和 T-Mobile 三家寡头，他们确实“赢”了行业，但作为股东究竟赢到了什么？ - **折旧与入账扭曲**：5 年期折旧意味着数据中心 80% 的成本在第一年不会直接反映在净利润上，掩盖了真实的资本消耗。 - **电信历史重演**：Verizon 和 AT&T 砸重金垄断了整个通信基础设施，但股票回报却陷入 25 年的长期停滞。 - **超大规模云厂商的 CapEx/Sales 比率**：吴恺（Kai Wu）指出，当前许多超大规模云厂商（Hyperscalers）的资本支出占营收比例，已经超过了互联网泡沫顶峰时期的 AT&T。 ### [54:35](https://www.youtube.com/watch?v=n889nI8sR84&t=3275s) 利润沉淀在何处：从苹果模式看 AI 护城河 面对这场重金砸出的“惨胜”（Pyrrhic victory），吴恺直言：如果原本做搜索业务可以轻松拿到 60% 的高利润率，为什么非要一头扎进这种重资产、资本密集型的泥潭？乔什更进一步指出了历史上真正的超级赢家——苹果（Apple）。在移动网络的大基建浪潮中，Sprint、T-Mobile、AT&T 和 Verizon 投入了天文数字，而苹果没有铺设一米网络电缆，却凭借在网络基础设施之上打造的顶级终端与 iOS 生态，席卷了整个行业绝大部分的商业利润。 问题不在于“资本支出是好是坏”，而在于“利润最终沉淀在哪一层”。吴恺强调，如果一家公司的护城河仅仅是“有更多钱可以砸进问题里”，那这根本算不上壁垒，资本密集型企业从投入资本回报率（ROIC）来看历来都是糟糕的投资标的。真正的护城河是网络效应和无形资产，就像当年无论投入多少资金也无法凭空克隆一个谷歌搜索一样。现场谈到如今各家企业面临的颠覆危机： - 微软（Microsoft）处在直面冲击的靶心； - 乔什坦言自己目前的许多日常搜索已经直接从 Anthropic 的 Claude 开始，谷歌搜索业务面临实打实的威胁； - 部分市场观点认为，真正的利润最终会沉淀在像 Palantir 这一类应用与服务层上——Palantir 站在底层 LLM 之上，作为赋能者教大企业如何落地这些技术并收取真金白银。 ### [56:16](https://www.youtube.com/watch?v=n889nI8sR84&t=3376s) 铁路繁荣的历史镜像：价值归于消费者与社会 迈克尔引用了莫·埃尔-埃利安（Mohamed El-Erian）或行业学者的一句论断：“由于激烈竞争，几乎所有的利润最终都将转移给终端消费者。”吴恺用 19 世纪的铁路建设潮作为对照：绝大多数修筑铁路的私营公司最终由于过度竞争和债务重压陷入破产，即使活到最后，也只能拿到类似公用事业般平庸的微薄利润。然而，铁路建设本身却创造了极其庞大的 GDP 与社会红利。 现场展示了来自 Azeem Azhar（《Exponential View》主笔）的一张经典图表： - **蓝线**展示了每年的铁路建设成本； - **绿线**展示了铁路公司的实际盈利水平（每逢金融恐慌便暴跌破产，平时也仅仅贴近成本线）； - **红线**则是铁路对整个国家 GDP 的巨大拉动。 乔什形象地总结：“真正赚到大钱的是谁？是在北卡罗来纳州生产家具、装进车厢运往另一个城市卖掉的厂商。”AI 的结局很可能如出一辙——它会极大地便利商业创建、降低全社会的生产门槛，但基础设施建造者本身却未必能把这部分价值锁定在自己的利润表里。 ### [57:37](https://www.youtube.com/watch?v=n889nI8sR84&t=3457s) 大模型的大宗商品化与“可口可乐”之争 超大规模云巨头通常会辩解称，他们不仅在建基础设施，还直接投资或深度结盟了底层 LLM（如微软与 OpenAI、亚马逊与 Anthropic），因而能分得一杯羹。但吴恺指出，LLM 模型层本身同样面临着被迅速“大宗商品化”（Commoditization）的严重风险。 回顾 2022 年 11 月 ChatGPT（基于 GPT-3.5）刚发布时，OpenAI 对竞争对手拥有高达一两个数量级的压倒性优势。而如今，大模型的发展演变成了一场激烈的“赛马”：行业领跑者的位置在 Gemini、Claude、GPT 之间反复易手；Meta 的 Llama 开源努力虽然步履蹒跚，但背后还有 DeepSeek 以及其他中国大模型的激进追赶。技术能力上的绝对代差已被抹平，行业竞争转移到了用户心智与渠道： | 角色 / 品牌 | 代表模型 | 市场定位与竞争优势 | |:--- |:--- |:--- | | **可口可乐 (Coke)** | **ChatGPT / OpenAI**（或被提议为 Gemini） | 家喻户晓的大众心智代名词；普通人提到 AI 第一反应就是它。 | | **百事可乐 (Pepsi)** | **Claude (Anthropic)** | 极受核心科技开发者与专业群体推崇，在推理能力上不断抢占头部。 | | **胡椒博士 (Dr. Pepper)** | 竞争梯队补充 | 维持多极竞争格局。 | | **隐形霸主 (分发之王)** | **Google Gemini** | 坐拥无与伦比的装机量与分发渠道（Gmail、安卓与原生搜索），每一笔搜索都在后台被 Gemini 化。 | 关于谁才是真正的“可口可乐”，乔什与迈克尔展开了风趣的争论：乔什认为即便普通人意识不到，谷歌在日常搜索和工作邮箱中强行内嵌 Gemini，其实际使用量可能已是全球第一；而吴恺则认为从技术评测上看 Gemini 目前可能屈居第三，下周若发布新版本胜负仍未可知。 ### [59:56](https://www.youtube.com/watch?v=n889nI8sR84&t=3596s) 转向无形资产：吴恺的量化投资溯源 讨论随后转向吴恺创立的投资机构 Sparkline Capital 及其核心方法论。迈克尔向吴恺抛出了一个两阶段的问题： 1. 究竟是什么契机促使吴恺将整个投资事业押注在“无形资产”（Intangible Assets）这一概念上？ 2. 当前科技巨头疯狂的 AI 硬件与基础设施资本支出，对量化投资组合中的无形资产评估产生了怎样的影响？ 吴恺回顾了自己的职业生涯：他曾就职于著名量化资产管理公司 GMO（与 AQR 并称量化界先驱）。在 GMO 等传统量化机构中，价值因子（Value Factor）是绝对的当家招牌——即系统性地买入市净率（Price-to-Book）便宜的股票，做空或低配昂贵的股票，这种方法在历史上曾带来丰厚的超额回报。然而，随着现代经济底层逻辑从工厂、土地等有形资产全面转向专利、品牌、算法及数据等无形资产，传统的市净率框架遭遇了前所未有的失效挑战，这直接催生了他对重新定义“无形价值”的执着探索。 ### [01:00:42](https://www.youtube.com/watch?v=n889nI8sR84&t=3642s) 传统价值投资的失效与无形资产重估 Kai Wu 分享了自己在量化研究员时期的经历，当时他的核心工作之一就是帮助团队更新和现代化捕捉价值因子的方法。他坦言，对于依靠传统系统化因子定义的价值投资者而言，过去这些年极其艰难。主持人提到马克·墨比尔斯（Mark Mobius）曾半开玩笑地指出一个戏剧性事实：经典的法玛-弗伦奇（Fama-French）价值因子——本质上是一个做多便宜股票、做空或低配昂贵股票的指数体系——在论文发表前的 80 年里年年赚钱，但自从 1994 年论文正式发表、或者说从美国在线（AOL）横空出世以及个人电脑普及时起，这个策略就基本不再灵验了。 大家探讨的核心并非在于“低买高卖”这一古老智慧失灵了，而是“内在价值”的锚定基准彻底发生了改变。本·格雷厄姆（Benjamin Graham）在一一个世纪前通过《证券分析》奠定的价值投资基石，在逻辑上无懈可击。然而，传统的会计指标严重滞后，仅仅聚焦于厂房、设备等有形资产，却将研发投入（R&D）、广告与营销支出以及人力资本直接当期费用化，完全无法适应从传统工业经济向以英伟达（Nvidia）、谷歌（Google）为代表的“轻资产、信息驱动型”现代经济的转型。Kai Wu 强调自己依然坚定地自视为一名价值投资者，因为拥有严谨的价格纪律、买入便宜货并在合理时机止盈，始终是市场的稳定器；但核心前提是必须更新衡量“真实价值”的指标。 沃伦·巴菲特（Warren Buffett）的投资演进便是最经典的例证。巴菲特起初是格雷厄姆最正统的门徒，早期甚至买下濒临破产的伯克希尔·哈撒韦纺织厂，专门搜寻所谓“烟蒂股”（格雷厄姆当年甚至专挑股价低于每股净现金的公司，而格雷厄姆晚年退休时也感叹“这套东西已经行不通了”）。直到遇到查理·芒格（Charlie Munger），巴菲特才发誓不再碰那些烂生意，转而买进可口可乐（Coca-Cola）和苹果（Apple）。数据显示，巴菲特一生中约有 84% 的投资标的买入时市净率（P/B）都大于 1，这证明他早就摒弃了简单的账面净资产估值框架，将无形资产护城河融入了自己的体系之中。 - 法玛-弗伦奇价值因子在 1994 年学术论文发表前拥有长达约 80 年的有效盈利期，但随着互联网与个人电脑普及，传统账面价值指标逐渐失效。 - 传统会计准则将研发费用、品牌营销和人力资本视为短期成本而非长期资本性资产，导致对轻资产科技企业的真实价值产生严重系统性低估。 - 统计显示巴菲特约 84% 的持仓买入时市净率大于 1，反映其投资范式早已从格雷厄姆式的有形资产清算价值转向无形护城河。 --- ### [01:03:53](https://www.youtube.com/watch?v=n889nI8sR84&t=3833s) 用大语言模型量化无形护城河 既然常识告诉我们品牌、核心技术与人力资本至关重要，为何量化投资界长期对此无能为力？Kai Wu 解释称，过去几十年量化基金被困在 CRSP 和 Compustat 这类传统结构化财务数据库中，依赖的核心工具仅仅是线性回归等几板斧。这套工具在面对无形资产时完全束手无策，因为无形资产绝大部分隐藏在非结构化数据中——包括专利文件、商标注册、公司 SEC 财报披露附注、行业新闻、分析师报告以及社交平台（如 Twitter/X）的海量讨论。 随着大型语言模型（LLM）与自然语言处理（NLP）技术的突破，海量的非结构化文本数据终于得以被高效解析与特征提取。Kai Wu 指出，这使得量化团队历史上第一次有机会将巴菲特式的主观定性直觉转化为系统化的量化因子。虽然完全复刻这种大师级直觉依然任重道远，但人工智能正在填补基本面定性判断与系统化模型之间的鸿沟。 ```mermaid flowchart TD accTitle: 传统量化与现代无形资产量化对比 accDescr: 展示从结构化数据到利用LLM解析非结构化数据的量化投资演进路径 subgraph Trad["传统价值量化（易失效）"] A["传统数据库 (Compustat/CRSP)"] --> B["结构化财报 (P/E, P/B)"] B --> C["线性回归模型"] C --> D["传统价值因子 (忽略无形资产)"] end subgraph Modern["现代无形资测量化（新范式）"] E["海量非结构化数据<br/>(专利/商标/财报附注/社交媒体)"] --> F["大语言模型与 NLP 处理"] F --> G["量化无形资产护城河<br/>(技术实力/品牌认知/人力资本)"] G --> H["系统化巴菲特式价值投资策略"] end ``` --- ### [01:05:43](https://www.youtube.com/watch?v=n889nI8sR84&t=3943s) 品牌价值的量化：从彼得·林奇到社交媒体监测 现场探讨转向了具体如何量化“品牌”这种虚无缥缈的资产。主持人指出，当前市场中耐克（Nike）和露露乐蒙（Lululemon）可以说是处境极为糟糕的代表，两家公司的股价自历史高点均已重挫约 76%。大家热议的核心问题在于：量化模型能否在股价腰斩之前整整一年，就通过数据敏锐地捕捉到品牌力正在衰竭的先行信号？ 谈及品牌调研，乔什·布朗（Josh Brown）分享了他与传奇基金经理彼得·林奇（Peter Lynch）做过一小时专访的趣事。林奇在当年依靠的是最朴素的“模拟时代”调研法：亲自走进超市，观察哪款产品被放在收银台的核心黄金位，哪款产品被挪到了无人问津的货架角落，或者直接询问身边的亲友是否还在持续购买某个品牌。而在数字化时代，这种调研正在被全面系统化。量化团队可以通过爬取 Instagram 等社交平台的数据，跟踪品牌提及的情感倾向是正面还是负面，以及讨论这些品牌的是否是真正的核心潮流人群。Kai Wu 同时提醒，跟踪这类信号必须防范数据操纵与干扰，正如企业会粉饰财务报表一样，品牌也可以花钱雇佣网红买热搜、造声势，但只要模型能够合理剔除这些噪音，依然能够先于股价变动捕捉到品牌生命周期的拐点。 --- ### [01:07:36](https://www.youtube.com/watch?v=n889nI8sR84&t=4056s) 预测市场在公司级量化投资中的潜力与现状 讨论尾声触及了迅速兴起的“预测市场”（Prediction Markets）能否作为全新的另类数据来源。乔什·布朗认为预测市场极具吸引力，因为它能将许多远远超出官方 8-K 财报公告范围的市场观点、甚至极度前沿的事件概率进行量化定价，蕴含着丰富的信息量。 Kai Wu 对此表示认可，认为预测市场确实是一项非常有趣且具备战略潜力的新兴数据源。但他坦承，自己目前的量化模型中还没有大量接入预测市场数据。最主要的原因在于，预测市场目前的交易品类多集中在宏观政策、地缘政治或选举等大事件上，而他目前管理的策略绝大部分工作都深耕于个股层面的微观无形资产挖掘，因此在公司基本面层面的适配应用仍处于初步探索阶段。 ### [01:08:02](https://www.youtube.com/watch?v=n889nI8sR84&t=4082s) 预测市场、另类数据与 Kai Wu 的无形资产 ETF 几位主持人接着讨论了预测市场在宏观预测中的价值——比如预测总统大选的胜者。虽然在体育博彩或加密货币领域这套机制往往显得噪音偏多，但若将其套用到具体的商业公司上，情况就变得十分微妙。例如，是否能有一个“耐克能否打响翻身仗”的预测合约？即便有，其流动性可能也相当有限。此时，Kai Wu 提出了自己经常参考的另类数据源：Google Trends（谷歌趋势）。通过监控大众在谷歌上检索的关键词热度（比如消费者对 Lululemon 旗下热门产品“Align 瑜伽裤”等单品的搜索趋势），能够直观捕捉到品牌势能的上升或衰减。 话题随即转到 Kai Wu 管理的两只 ETF 产品上。Josh Brown 顺势起哄，让 Kai 当场向坐在对面的 Ben Carlson 推销：“他本人就坐在你面前，快说服他买！”Kai 介绍称，他在 2021 年推出了首只专注于美股的无形资产 ETF；随后推出的第二只则是聚焦发达国家市场的国际无形资产 ETF。两只基金遵循完全相同的量化策略：寻找相对于其“扩展内在价值”而言被低估的股票，而这种内在价值特别将无形护城河纳入考量，包括品牌声誉、人力资本、知识产权以及网络效应。 - **美股无形资产 ETF**：Kai Wu 于 2021 年成立，聚焦于美国本土具备强大无形护城河且估值合理的标的。 - **发达市场国际无形资产 ETF**：采用相同策略，将投资宇宙延伸至海外发达市场。 - **扩展内在价值体系**：突破传统有形资产账面价值的局限，深度量化品牌、人力资本、IP 以及网络效应等无形壁垒。 --- ### [01:09:23](https://www.youtube.com/watch?v=n889nI8sR84&t=4163s) 巨型独角兽闪电入指：SpaceX 与马斯克这个“终极无形资产” Josh 提出了一个当下指数生态中的新变局：如今科技独角兽上市后进入主要指数的速度比以往任何时候都要快。未来可能会出现像 SpaceX 这样的大体量公司，在 IPO 仅仅 10 天后就火速被纳入标普 500 指数，类似的情况也极有可能发生在 Anthropic 身上。面对这种新生态，量化选股模型该如何应对？Kai 表示，由于自己的基金属于绝对收益型基金，并没有强制跟踪的基准指数，因此“不需要为了被动配置而持有任何东西”。虽然 SpaceX 一旦入指便顺理成章进入了可投资宇宙，但他高度怀疑这类标的会在上市第一天就被纳入模型组合。 在谈及模型运作机制时，Kai 强调整个选股系统是“完全基于规则的量化模型”（Completely rules based）。但如果模型输出的结果出现违背基本面直觉的异常情况，他会反思底层逻辑是否有遗漏，并系统性地微调模型，以期在未来持续生效，而非单纯做主观特例干预。关于 SpaceX、Anduril 这类新锐巨头究竟算有形资产还是无形资产，Kai 与主持人们达成了高度一致：它们表面上是发射实体火箭或造防务硬件的物理公司，但绝大部分真实价值都源自知识产权、工程团队与前沿技术；而在 SpaceX 的估值中，甚至很大一部分就是伊隆·马斯克本人。Josh 随即打趣总结道：“伊隆·马斯克的深度参与，本身就是终极的无形资产。” ```mermaid flowchart TD accTitle: 无形资产评估体系与独角兽纳入逻辑 accDescr: 展示量化模型对公司扩展内在价值的定义，以及物理资产与终极无形资产的结合逻辑。 subgraph Valuation["扩展内在价值体系"] A["物理与财务资产"] --> C["综合内在价值评估"] B["无形护城河: 品牌/IP/网络效应"] --> C C --> D["严格量化规则筛选"] end subgraph Frontier["前沿硬科技公司 (如 SpaceX)"] E["物理实体: 火箭/制造发射场"] F["顶级工程团队与核心 IP"] G["关键人效应: 终极无形资产 (Elon Musk)"] E & F & G --> H["深度混合型高估值资产"] end D -.-> I["绝对收益投资组合 (不盲目跟风入指)"] H -.-> I ``` --- ### [01:10:52](https://www.youtube.com/watch?v=n889nI8sR84&t=4252s) Ben Carlson 的新书《Risk and Reward》：图表狂魔与被挤下封面的荐语 在一阵欢笑声中，谈话硬切到了 Ben Carlson 即将出版的新书《Risk and Reward》。由于不受基金合规禁言条例的限制，Ben 终于可以畅所欲言。Ben 坦言自己根本不擅长自我推销和搞销售，而 Josh 则是天生的顶级销售员——Michael Batnick 开玩笑形容 Josh 的销售天赋：“就算是在华氏 90 度的高温天，Josh 也能把番茄酱冰棍成功卖给一个戴着白手套的女士。” 自我调侃归自我调侃，Ben 还是为新书做了一次硬核推销：“市面上绝没有任何一本书的图表和数据表格能比这本更丰富！”全书在专业图表团队 Chartr 的协助下，整整收录了 52 张深度图表与数据表，几乎每一章都用数据说话。面对主持人关于“满是图表的书听有声书该怎么办”的戏谑，Ben 解释称购买有声书的听众会收到包含全书所有图表的配套 PDF。 期间现场还爆出了一个装帧幕后花絮：好友 Nick Maggiulli 原本为新书写了推荐语，并排在封面位置；结果出版社为了版面效果，临时把 Nick 的推荐语挤到了内封，换上了畅销书作家 Morgan Housel 的大牌荐词（“Ben Carlson 已经精通了向投资者揭示最关键少数大议题的艺术”）。大家翻着样书确认 Nick 确实被“降级”后，爆发出一阵充满同情的欢笑。 --- ### [01:12:49](https://www.youtube.com/watch?v=n889nI8sR84&t=4369s) 市场周期的“倍速播放”：暴跌压缩成急速回撤？ 聊到书中的第 50 号图表时，Ben 展开了一个核心论点：他统计了美股历史上表现最差的单日、单月和单年。在历史上，标普大盘单月跌幅达到或超过 20% 的极端情况总共只出现过 6 次，而且几乎全部集中在 1930 年代、1940 年代以及 1987 年的“黑色星期一”。 Ben 提出了一个对未来市场极具颠覆性的假设：过去那种历时两三年、阴跌 60% 的漫长熊市危机可能正在成为历史；取而代之的是更加急速凶狠的“空气悬崖”（Air pockets）。在当今信息瞬时传播、算法高频博弈以及全天候 24 小时交易的环境下，市场消化坏消息的速度被无限拉快，修正周期正被“倍速通关”（Speed running the correction）。现代财政和货币政策的极速介入基本切断了像 1930 年代“大萧条”那样的长期极左侧尾部风险，但风险本身不会凭空消失，其转化结果就是跌幅高度压缩在短短一两个月内——“可能短短一个月内就瞬间暴跌 25%”。 对此，大家展开了热烈探讨。如果周期的齿轮注定越转越快，加上 AI 对信息差的进一步抹平，传统“买入并持有”策略的心理门槛实际上变得更加残酷，对人性抗压能力的考验不仅没有减轻，反而以更短促、更剧烈的形式轰炸着每一个盯盘的投资者。 - **单月暴跌 20% 以上的极罕见历史**：美股历史上仅发生过 6 次，大多发生在 1930 年代大萧条时期及 1987 年崩盘。 - **危机模式演变**：从“耗时数年阴跌 60%”逐渐演变成“一至两个月内闪崩 25%”的急剧下坠。 - **尾部切除后的风险守恒**：政府财政与美联储货币干预阻断了大萧条重演，但波动性以更加高频、瞬时的速度集中宣泄。 --- ### [01:14:32](https://www.youtube.com/watch?v=n889nI8sR84&t=4472s) 日本股市的 53 号图表：长跑 60 年的均值回归真相 在访谈尾声，Ben 拿出了全书引发争议最多的第 53 号图表——关于日本股市（日经指数）的长期真相。在投资圈，每当长期主义者倡导“长期持有股票必有回报”时，反对者最常用的反驳利器永远是：“那日本呢？日本人可是守着阴跌不动的股市整整套牢了三十年！” Ben 澄清了这一被断章取义的认知误区。表面上看，日本股市在 1989 年 12 月见顶，随后一路震荡阴跌直到 2009 年 3 月才彻底见底；从 1990 年到 2024 年这三十多年间，日股的年化复合回报率确实仅有可怜的 1.5% 左右。然而，如果把时间轴前移拉长至整整 60 年（从 1970 年统计至 2024 年），投资日本股市的年化复合回报率依然接近惊人的 9%。 这背后的根本逻辑在于**收益率的前置高度压缩**： 1. **1970 年至 1989 年的大繁荣**：在这前二十年里，日本整体股市年化回报率高达 22%，而日本小型股的年复合增长率甚至达到了不可思议的 30%。 2. **均值回归的代价**：二十年透支了半个世纪的繁荣，估值泡沫极度吹大后，必然面临接下来三十余年每年仅 1% 的长期均值回归消化期。 Ben 指出，虽然长达三十余年的停滞在个体的生命周期里确实漫长得令人绝望，但如果拉长到跨越半个多世纪的完整大周期来看，“长期投资在日股依然起到了作用”，只是它以极其极端的方式向世人展示了周期均值回归的铁律。 ### [01:15:43](https://www.youtube.com/watch?v=n889nI8sR84&t=4543s) Ben Carlson 的新书《Risk and Reward》：历史大跌背后的乐观主义 在讨论完市场周期后，主持人将话题引向了 Ben Carlson 的新书《Risk and Reward》（风险与回报）。Josh Brown 与 Michael Batnick 在现场打趣，调侃大家是打算通过纸质书阅读还是听有声书入睡。Josh 直言自己是“老派”读者，已经迫不及待看完了前两章，并盛赞 Ben 的文字是当前市场最需要的一剂“良药”（"This is what I need. This is the medicine."）。 Josh 指出，市面上绝大多数金融书籍都在兜售末日恐慌——要么宣称美元即将失去全球储备货币地位，要么鼓吹黄金将全面替代法币。而 Ben 的书截然相反，其核心逻辑建立在坚定的长期乐观主义之上，即“一切大概率都会变好”（"things will probably be okay"）。书中用扎实的历史数据证明：美股在过去 100 年里实现了年化约 10% 的长期回报，而这一数据已经完全涵盖了历史上所有的极端危机。 - **危机已被定价**：过去百年年化 10% 的收益并非在风平浪静中取得，它已经完整吸收了 1929 年大萧条（The Great Depression）、1987 年的“黑色星期一”（Black Monday）以及 1970 年代的恶性滞胀期。 - **承受波动的回报**：长期回报本身就已经计入了市场所遭遇的“所有糟糕破事”（"inclusive of all the bad shit that's happened"），这正是长期投资获得丰厚回报的根本逻辑所在。 ```mermaid flowchart LR accTitle: Long-term Market Resilience accDescr: Diagram showing how historical market crises are absorbed within long-term returns. subgraph Crises["历史极端冲击 (已全数计入)"] A["1929年 大萧条"] B["1970年代 恶性滞胀"] C["1987年 黑色星期一"] end subgraph Reality["市场韧性与长期结果"] D["过去100年年化约 10% 回报"] E["'Things will probably be okay'"] end Crises --> D D --> E ``` ### [01:17:02](https://www.youtube.com/watch?v=n889nI8sR84&t=4622s) 致谢嘉宾 Kai Wu 与节目尾声 节目的最后部分进入总结与致谢环节。Josh Brown 对嘉宾 Kai Wu 带来的深度分享表达了由衷赞赏，称赞他“不仅管理着优秀的基金，更是一位用硬核数据验证投资假说的思想家与哲学家”。现场还开起玩笑，说 Kai 实在太有深度，平时“根本不屑于写博客，一写就是学术级别的白皮书（white papers）”。 Kai 透露，想要跟踪他最新研究与观点的观众可以访问其公司官网 sparklinecapital.com，或者在社交平台 X（Twitter）上关注他的动态。最后，Josh 向本期节目的幕后制作团队（John、Duncan、Rob、Nicole、Daniel、Travis 等）致以谢意，并在全场掌声中为团队成员 Graham Thomas 庆祝生日，正式结束了本期节目的录制。 - **嘉宾研究去向**：Sparkline Capital 官网（sparklinecapital.com）。 - **幕后制作人员鸣谢**：John、Duncan、Rob、Nicole、Daniel、Travis 等团队成员，并向 Graham Thomas 送上生日祝福。 ## Notable quotes > [02:36](https://www.youtube.com/watch?v=n889nI8sR84&t=156s) The market is so horny right now — Downtown Josh Brown > [24:18](https://www.youtube.com/watch?v=n889nI8sR84&t=1458s) Guys, let's cut to the chase. Do we do an AI? — Downtown Josh Brown > [29:47](https://www.youtube.com/watch?v=n889nI8sR84&t=1787s) AI 技术本身极具变革性，但它在当前周期里依然可能是一笔糟糕的投资。 — Kai Wu > [32:08](https://www.youtube.com/watch?v=n889nI8sR84&t=1928s) 宁可过度投资破产，也绝不能在这场竞赛中落后 — Mark Zuckerberg > [40:15](https://www.youtube.com/watch?v=n889nI8sR84&t=2415s) the most logical melt-up of all time — Michael Batnick > [56:16](https://www.youtube.com/watch?v=n889nI8sR84&t=3376s) 真正赚到大钱的是谁？是在北卡罗来纳州生产家具、装进车厢运往另一个城市卖掉的厂商。 — Downtown Josh Brown > [01:09:23](https://www.youtube.com/watch?v=n889nI8sR84&t=4163s) 伊隆·马斯克的深度参与，本身就是终极的无形资产。 — Downtown Josh Brown > [01:10:52](https://www.youtube.com/watch?v=n889nI8sR84&t=4252s) 就算是在华氏 90 度的高温天，Josh 也能把番茄酱冰棍成功卖给一个戴着白手套的女士。 — Michael Batnick > [01:10:52](https://www.youtube.com/watch?v=n889nI8sR84&t=4252s) 市面上绝没有任何一本书的图表和数据表格能比这本更丰富！ — Ben Carlson > [01:10:52](https://www.youtube.com/watch?v=n889nI8sR84&t=4252s) Ben Carlson 已经精通了向投资者揭示最关键少数大议题的艺术 — Morgan Housel > [01:15:43](https://www.youtube.com/watch?v=n889nI8sR84&t=4543s) This is what I need. This is the medicine. — Downtown Josh Brown > [01:15:43](https://www.youtube.com/watch?v=n889nI8sR84&t=4543s) things will probably be okay — Downtown Josh Brown > [01:15:43](https://www.youtube.com/watch?v=n889nI8sR84&t=4543s) inclusive of all the bad shit that's happened — Downtown Josh Brown ## Mentioned Downtown Josh Brown, Michael Batnick, Ben Carlson, Kai Wu, Hulk Hogan, Sylvester Stallone, Vince McMahon, Sam Altman, Patrick O'Shaughnessy, Krishna Rao, Jeremy Grantham, Edward Chancellor, John Templeton, Mark Zuckerberg, Jeff Bezos, Jason Goepfert, Dan Greenhaus, Peter Lynch, Benjamin Graham, Charlie Munger, Warren Buffett, Elon Musk, Nick Maggiulli, Morgan Housel, Graham Thomas, Anthropic, Cerebras Systems, Nvidia, OpenAI, Oracle, Betterment, Sparkline Capital, Snowflake, Google, Amazon, Microsoft, Meta, Salesforce, Adobe, BlackRock, Digital Realty Trust, Micron Technology, AT&T, Verizon, T-Mobile, Apple, Palantir, SpaceX, Anduril, Claude, Claude Code, Claude 3.5 Sonnet, Opus, ChatGPT, Gemini, Llama, DeepSeek, Risk and Reward, Animal Spirits, A Wealth of Common Sense, ITAN, DTAN, SMH, SOX, Chartr ## Video description On episode 242 of The Compound and Friends, ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Michael Batnick⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ and ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠Downtown Josh Brown⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ are joined by Ben Carlson and Kai Wu for a wide-ranging conversation on the AI boom, Nvidia’s massive run, Anthropic’s explosive growth, the Cerebras IPO, software stocks getting crushed, and whether today’s tech giants are becoming the new utilities. They also discuss the capital cycle behind AI infrastructure, where value might actually accrue, the risks of market concentration, why intangible assets matter more than ever, and Ben’s new book Risk and Reward. This episode is sponsored by: Betterment Advisor Solutions. To learn more, visit ⁠https://www.betterment.com/advisors Buy Ben's new book: https://lnk.to/riskandrewardbook ►00:00 - Cold Open ►12:17 - The AI Trade ►34:08 - Echos from the Past (Are software stocks in trouble?) ►40:12 - Nasdaq Euphoria ►45:44 - The Biggest Risk to the Market ►52:40 - Mag 7 Capex and Re-rating Risk ►59:56 - Kai Wu on Intangible Value ►01:07:32 - Prediction Markets ►01:10:51 - Ben's New Book, Risk & Reward 👕 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 #AI #Nvidia #Anthropic #Markets #Finance [Source](https://www.youtube.com/watch?v=n889nI8sR84)

[2:36] Downtown Josh Brown: The market is so horny right now [24:18] Downtown Josh Brown: Guys, let's cut to the chase. Do we do an AI? [29:47] Kai Wu: AI 技术本身极具变革性，但它在当前周期里依然可能是一笔糟糕的投资。 [32:08] Mark Zuckerberg: 宁可过度投资破产，也绝不能在这场竞赛中落后 [40:15] Michael Batnick: the most logical melt-up of all time [56:16] Downtown Josh Brown: 真正赚到大钱的是谁？是在北卡罗来纳州生产家具、装进车厢运往另一个城市卖掉的厂商。 [69:23] Downtown Josh Brown: 伊隆·马斯克的深度参与，本身就是终极的无形资产。 [70:52] Michael Batnick: 就算是在华氏 90 度的高温天，Josh 也能把番茄酱冰棍成功卖给一个戴着白手套的女士。 [70:52] Ben Carlson: 市面上绝没有任何一本书的图表和数据表格能比这本更丰富！ [70:52] Morgan Housel: Ben Carlson 已经精通了向投资者揭示最关键少数大议题的艺术 [75:43] Downtown Josh Brown: This is what I need. This is the medicine. [75:43] Downtown Josh Brown: things will probably be okay [75:43] Downtown Josh Brown: inclusive of all the bad shit that's happened

[12:17] Josh Brown: I really don't think there's anything good going on in the economy other than the AI CapEx boom. [19:23] Josh Brown: In 7 trading days, Nvidia has added $900 billion in market cap, which is as large as McDonald's, Disney, Boeing, Uber, Starbucks, and Royal Caribbean combined... These numbers almost don't mean anything anymore. [24:18] Kai Wu: Whenever a new technology comes out, you want to start by giving it low-end tasks where mistakes are forgivable... At first AI was a good analyst. Now, as a quant... Claude Code or Codex are perfectly capable of doing a lot of that analysis. [29:26] Kai Wu: two things can be true at once: AI technology can be transformative and it may also be a bad investment. [33:05] Kai Wu: If Sam Altman and OpenAI get AGI and I'm not even doing anything, I'm done... so I have to compete. Once Sam goes all in, everyone else has to go all in. [36:25] Josh Brown: I am starting to believe that they are newspapers... A lot of publicly traded software companies are going to be zeros—not like zero zero, but never coming back. [46:35] Josh Brown: Buy stocks... Just don't let the damn robots.

[0:00] All right. Josh, didn't you read your [0:01] own audio book? It's an amazing show. [0:03] Yeah, of course. Can you imagine [0:04] somebody else reading my book? [0:05] >> Didn't your brain feel like mush after [0:07] doing that though? I did I did it. It [0:09] took me 6 weeks. [0:10] 6 weeks? Yeah, see I did it in 2 days. [0:13] Cuz I I uh Can you all switch [0:15] headphones? Oh, sure. [0:17] I don't know if you heard any of my [0:18] audio book, but I really Yeah, why are [0:20] on I performed that [ __ ] thing like [0:22] Mick Jagger. [0:23] >> [laughter] [0:24] >> Like I was like [0:25] I'm like yelling into the microphone. [0:28] Takes a lot It's a performance. It takes [0:29] a lot of energy. [0:30] >> Yes, it is. Yeah, I was dead afterwards. [0:31] >> Mine is so autobiographical it could [0:33] never be read by somebody else. It would [0:35] sound ridiculous. [0:37] You know what I mean? I actually enjoyed [0:38] the process, but it was a lot. I hated [0:40] it. [0:41] >> [laughter] [0:41] >> I like how it came out. I like how it [0:44] came out. Yeah, I enjoyed it after it [0:45] was done. [0:47] >> I like I'm happy with the product, but I [0:49] I don't know if I want to do it again. [0:53] You working on your next book yet or [0:55] >> [applause] [0:55] >> I need a break. [0:56] >> Any notebooks for you? For myself. Not [0:58] yet, but I kind of want to do one now. [1:00] Well, all your white papers, put them [1:01] all together, that's a book. [1:02] There you go. [1:03] >> spoken to him or has he spoken to you? [1:04] Has he reached out to you? I should talk [1:05] to him. [1:06] >> Yeah, yeah. Why do you want to Why [1:09] devil's advocate, why do you want to do [1:10] a book? [1:11] Cuz all the cool kids have one. [1:13] Okay. Do you think people read books? [1:17] No, but it's a it's a cool thing to have [1:18] done. He just he His book got published [1:20] yesterday. Why are you talking [ __ ] No, [1:22] I'm not I've written four books. I'm not [1:24] talking [ __ ] Dude, I'm reading his book [1:26] right now. Do you think people read [1:28] books? [1:28] >> I don't think people read books anymore. [1:30] I don't know. [1:30] >> So, you did a lot of podcasts for yours. [1:33] The percentage of people who asked the [1:34] Okay, I did that. [1:36] >> The percentage of people who I'm on a [1:37] podcast with who have read my book is [1:39] probably 10%. I can tell if they read [1:41] the book or not. [1:42] Yeah, no I don't think anybody reads. [1:44] >> And I don't I don't I don't like fault [1:45] them for that. I don't even think people [1:47] read articles. I think people read [1:48] headlines. [1:50] And soon they're only going to watch [1:52] videos about articles. I read so many [1:54] headlines today. I'm exhausted. [1:57] I love I mean I Look, I I love reading. [2:00] I just I don't know how many people read [2:02] are reading books these days. I'm not [2:04] sure. I've seen surveys on it. [2:06] >> books. It's not the same thing. [2:07] >> Yes. People buy books [2:10] cuz they want to read them. They truly [2:12] want to. [2:13] But I think our brains have been rewired [2:15] by algorithms. [2:16] >> This is why audiobooks are so great. [2:18] Yes, I agree. You can have them on in [2:20] the background. [2:21] >> He's an audiobook. Michael's an [2:22] audiobook. [2:22] >> it. I've never seen a bigger audiobook [2:24] guy than me. You know what? Listen, this [2:25] is your [laughter] book. What if I [2:26] listen to your book? [2:27] >> Yes, it's like I'm talking to you. It'll [2:29] put you to sleep at night. [2:30] I'm going to press pause and interrupt [2:32] you. Put your Put your sleeping mask on [2:34] and [2:36] By the way, someone like I Michael's got [2:37] a nice shirt on. Someone last night [2:38] said, [2:39] "Hey, I've been noticing Michael's [2:41] dressing better. Did Did you start [2:42] dressing him?" [2:43] >> He's so He's so rich now. That's what's [2:45] going on. Wow. Appreciate that. He is [2:48] dressing better. He is. That's a [2:50] compliment to say that I dress you? [2:51] >> It's not just that he's dressing better. [2:52] His taste is improving. It's cuz cuz you [2:55] could have bad taste and buy a lot of [2:57] expensive clothes that don't look good. [2:58] He's like put together. Like when you [3:00] see him at an event [3:02] >> fashionable. He No, he is. [3:03] >> Who is this guy? I know. He is. Who is [3:05] this guy? This guy. He's so liquid right [3:07] now. [3:08] It's a bull market. He's so liquid. You [3:09] can't trust nothing in a bull market, [3:11] Mike guy? That's right. [3:13] All right. [3:15] So [3:15] >> [clears throat] [3:15] >> So guys, this is going to be a good one. [3:16] The market is so horny right now. [3:20] >> [laughter] [3:21] >> Full on. A full I mean, the market is [3:23] just Did you watch DTF? [3:25] Yes. [3:26] How [3:27] >> It's It's I liked it. It's slow though. [3:28] It's so good. No, it's a slow burn. I I [3:30] did like it. I didn't see the whole [3:32] thing. [3:32] >> you got the full on reference? Yeah, [3:33] yeah, yeah. I I just uh [3:35] >> it, Ben? Weirdest show ever. [3:36] >> You That That little smirk of yours [3:38] makes me feel like you watched it. [3:39] >> I liked that it was weird. It's not what [3:41] I expected. It's definitely not what I [3:43] expected. [3:43] >> They're so When they're just talking to [3:44] each other, just signing, like just [3:46] Yeah. [3:47] What's that actor's name? Uh the [3:48] Stranger Things guy? [3:50] David Harbour. David Harbour. Big Knicks [3:52] fan. He's good. Yeah. [3:54] All right. I like that. [3:55] >> [applause] [3:56] >> I watched the whole Hogan documentary. [3:59] Uh Phenomenal. This this week. That's [4:01] what I mean. [4:02] >> That's what That was a very good one. [4:03] Netflix, four episodes. [4:05] It was so good. Because I think because [4:08] it's like [4:10] So my childhood, I was like, I don't [4:12] know, 8 years old for WrestleMania. [4:14] So like that was it for us. That's the [4:16] only thing we cared about. [4:17] >> Silverdome in Detroit? How many guys or [4:20] how many people [4:21] in the last [4:23] 75 years [4:25] do you think were on the top 10 [4:28] globally most famous like most [4:30] recognized name list? [4:31] >> 10 [4:32] There's only been 10 for the last 75 [4:34] years. [4:34] >> how many people are on the top 10? [4:35] >> saying it's a Shut up, [ __ ] It's a [4:37] small list. It's a list that doesn't [4:40] turn over a lot. No, no, no. No, no, no, [4:42] no. He's one of the most famous athletes [4:43] who ever lived. People, 100%. [4:45] >> People. Yeah, maybe people. One of the [4:47] most instantly recognizable people. [4:50] >> Right. So [4:52] So there's that part of it. Also [4:55] he had not like a rise and fall, he had [4:58] a rise, a fall, a rise, a fall. It's [5:00] like maybe four or five cycles of [5:03] everyone loves him, everyone hates him. [5:05] >> It was cinema. It I mean, it's a really [5:07] It's a really great doc and it was very [5:09] well done. [5:09] >> in Rocky III. And he inspired me to get [5:11] some [5:12] exercise. [5:13] >> Vince McMahon told him, "If you go to [5:15] Rocky III, you're fired." [5:18] Like he's like, "No, no, no, you have to [5:20] be at an autograph signing somewhere." [5:22] And Hulk Hogan's like, Oh yeah, brother. [5:24] Yeah. Let me tell you something, [5:25] brother. That He's like, "No, I have to [5:27] go to LA. Sylvester Stallone called me [5:28] and said I could be in the movie." [5:30] Thunderlips. [5:31] >> Vince Vince Vince said, "If you go, [5:33] you're fi- you're fired." Can you [5:34] imagine? Good luck, yeah. Yeah, good [5:36] luck with that. So I I thought it was uh [5:38] I thought it was really well done. Um [5:40] big big recommendation. [5:42] So [5:43] What else do you What What else do you [5:44] watch? [5:45] >> I want a podcast. [5:47] Yeah. [5:47] >> [laughter] [5:48] >> Wait, what else what else are you [5:50] watching right now? Uh I need a new one. [5:52] Playoffs. I'm watching basketball. Um [5:54] shows [5:55] There's a new show on Apple [5:57] um with the guy from Matt Reese. What's [6:00] it called? [6:01] Anybody else watching that? [6:03] Uh [6:03] >> Oh, the guy from The Americans. Yeah. [6:05] Yeah, I'm watching. What is it called? [6:06] >> Widows Bay. Widows Bay. What is it [6:08] about? [6:09] It's hard to it it's it's basically [6:11] they're on a they're on a Nantucket like [6:13] island [6:14] off the coast of uh Massachusetts and if [6:16] you leave the island it's like stupid [6:17] it's like [6:18] >> Hogan in it? It's a little Shutter [6:20] Island. It's not a totally out. [6:21] >> No Hulk. No Hulk. No Hulk. Oh, uh [6:24] Margot's got money troubles. Did you try [6:26] that? [6:27] >> Mhm. I watched the first one. [6:29] It's it's uh [6:31] Don't watch it with kids. Oh. I I I [6:33] won't. [6:34] >> I've been telling this to Ben. Uh I I've [6:36] been trying to watch Your Friends & [6:37] Neighbors. I'm on episode two. [6:40] My kids are cuz I Kobe's going to sleep [6:42] later. He's like going to sleep like [6:43] 9:00 9:15. [6:44] >> Can't watch that with kids. [6:46] Obviously. [6:46] >> That's No, the point is I'm sleeping by [6:48] 9:45. I just don't have the bandwidth. I [6:50] just go straight to sleep. Can't stay [6:51] up. [6:51] >> That's a good one. That's a good one. [6:52] Did you watch that? Your Friends & [6:54] Neighbors that's a good one. It's rich [6:55] people in Connecticut. [6:56] Yeah, and now they're on season two. [6:59] That's a That's a good one. Jon Hamm and [7:00] Anna Kendrick. [7:01] >> The reason why I like this show I was [7:02] telling Ben even though these characters [7:04] are obviously so unrelatable in terms of [7:06] the way that they live their life. Like [7:07] I don't know anybody like that. [7:09] The characters themselves are relatable [7:11] just in terms of like their the human [7:12] element of these people [7:13] >> [clears throat] [7:13] >> even though they're you know [7:14] gazillionaires or whatever. [7:17] They they talk about real [ __ ] [7:19] >> [laughter] [7:20] >> God damn it Kai, say something. [7:23] Michael's trying to get you to chime in. [7:24] Are you a non-TV person? [7:25] >> I have a one-year-old at home. I watch [7:27] two shows, Cocomelon and Sesame Street. [7:29] >> There we go. So unless you want to talk [7:31] about Elmo Cocomelon is slop. Oh my god. [7:34] It it goes in your head [7:36] over and over and over again every day, [7:38] right? Every day. [7:39] >> Cuz they want to see the the thing over [7:40] and over again. At one point I got my [7:41] daughter into K-pop Demon Hunter, which [7:43] actually has decent music. Great great [7:45] music. But that was a that was a fad and [7:46] she kind of got over it. Now she's back [7:47] into Cocomelon. So, is she walking? She [7:49] is, yeah. Okay, she's a good [7:50] >> She's a very good walker. [7:51] >> Oh, that song from K-pop Demon Hunters [7:53] could drive a a a a grown man to the [7:55] brink of insanity. I like it. [7:58] Like golden? Let's talk intangibles. [7:59] Yeah. All right, now you got me. [8:02] Return to Yharnam. Now I'm here, [8:04] powering up. You started it up? All [8:06] right, let's go, Johnny. All righty. [8:07] >> Do it. [8:08] >> How about a friends? Episode 242. [8:10] [applause] [8:21] >> [music] [8:22] >> What growth strategy are leading RIAs [8:24] using that most firms don't? [8:26] Segmentation. [8:27] >> [music] [8:27] >> Some clients' needs are sophisticated [8:29] and require deep ongoing planning. Some [8:32] clients' needs are simple, like those in [8:33] the wealth accumulation stage. [8:35] >> [music] [8:35] >> The smartest firms know planning [8:36] shouldn't look the same for every [8:37] client, but the experience should always [8:39] be exceptional. [8:40] >> [music] [8:40] >> Now it can be with Betterment Advisor [8:42] Solutions. It's the platform built for [8:44] segmenting your book and streamlining [8:46] these smaller and simpler [music] [8:46] accounts. The onboarding experience is [8:48] automated and paperless. The portfolio [8:50] management is streamlined and [music] [8:51] tax efficient. The client experience is [8:54] consistent and modern, and the impact [8:56] isn't just felt by your clients, it's [8:57] felt across [music] your entire [8:59] practice. Imagine a back office that's [9:01] humming, a team that's thriving, and a [9:02] service [singing] model ready to scale. [9:04] Betterment Advisor Solutions, your [9:06] biggest regret will be not doing it [9:07] sooner. Learn more at [9:08] betterment.com/advisors. [music] [9:11] All right. [9:14] Episode 242 of the number one investing [9:16] podcast on planet Earth. [9:19] My name is Downtown Josh Brown. [9:20] First-time listeners, first-time [9:22] viewers, thank you for joining us. [9:24] This handsome gentleman, well-dressed [9:26] gentleman to my left, [9:28] is Mr. Michael Batnick, co-host of the [9:30] show. [9:31] Say hello, Michael. [9:32] >> Hello, hello. [9:33] All right. [9:33] >> [applause] [9:35] [cheering] [9:35] >> We have two returning guests and I'm [9:36] super for this episode because we are [9:39] [music] in a market that is literally on [9:41] fire right now. [9:43] We have so much to talk about. Ben [9:45] Carlson needs almost no introduction. [9:48] [music] [9:49] Uh Ben is the head of [9:52] Yeah, do this do this stupid point. [9:57] >> [laughter] [9:59] >> All right. Benny uh Ben needs almost no [10:01] introduction, but we'll give him one. [10:02] Ben is the head of institutional [music] [10:03] asset management at Ritholtz Wealth. He [10:06] is the co-host of the wildly popular [10:09] Animal Spirits podcast with Michael [10:11] Batnick. How long has that been running? [10:14] We started in November. [10:15] >> November 2017. [10:16] >> 2017, yeah. [10:17] >> Oh my god. How the time flies. [10:20] Um and he is the author of a Wealth of [10:21] Common Sense, which is a blog that for [10:23] more than a decade has been required [10:26] reading for anyone serious about [10:28] investing. Ben is a CFA. He spent his [10:30] entire career managing money for [10:32] endowments, foundations, and long-term [10:35] investors. He's also the author of the [10:37] brand new book Risk and Reward, which [10:39] we're going to dig into today. [10:40] Congratulations, Ben. [10:42] No one will read it. No, I'm just [10:44] kidding. Uh [10:45] this is a huge accomplishment for you [10:46] because not just because of how prolific [10:49] you are in terms of publishing books, [10:51] but the blog you're three three days a [10:54] week? Ish? The blog helps me write the [10:56] book. I So that I totally understand as [10:59] a blogger turned author, but still it's [11:01] a it's a lot of work. It's very [11:03] impressive. Congratulations. [11:06] All [applause] right. [11:07] And another returning guest, Kai Wu. Kai [11:10] is the founder and chief [11:14] >> [applause] [11:15] >> founder and chief investment officer of [11:17] Sparkline Capital, an investment [11:19] management firm applying [11:20] state-of-the-art machine learning and [11:23] computing to uncover alpha in large [11:26] unstructured data sets. You have two [11:29] ETFs. Yes. [11:31] >> say so I I don't know if the rules allow [11:32] us to say. We're allowed to say. [11:33] >> We could say it but you can't confirm or [11:34] deny that they exist. Is that how it [11:36] works? [11:36] >> [laughter] [11:37] >> Somebody told me the ETF [11:39] compliance rules literally the most [11:41] insane [ __ ] I've ever seen. So you can't [11:44] talk about your own ETF on your own [11:47] podcasts or blogs but if somebody [11:50] interviews you for their magazine or [11:54] website or whatever then you can engage. [11:56] So I can ask a question. I can respond [11:58] to your question. So Kai, you are the [12:00] founder of Sparkline and you have two [12:01] ETFs I10 and D10. Can you confirm or [12:03] deny? Do not say anything until I check [12:05] if you can. [12:07] Okay. [12:08] Okay. And you must answer in the form of [12:11] interpretive dance. [12:12] So I don't know if you want to say. All [12:15] right. I [12:16] >> [applause] [12:18] >> The AI trade came back to the front [12:20] burner very quickly after liberation [12:23] day. It basically became the only game [12:24] in town. [12:26] I said on TV today I really don't think [12:29] there's anything good going on in the [12:30] economy other than the AI CapEx boom. I [12:33] think it's basically taken over. [12:36] Anywhere that you see economic growth [12:37] you could trace it back to something [12:39] that has to do with a trillion dollars [12:41] in CapEx spending directly related to [12:44] this AI buildout and everything else is [12:46] kind of boring and or bad. But I don't [12:49] want to gloss over the fact that the AI [12:51] trade was dying. [12:53] Like the Oracle blowup as a result of [12:56] what Sam Altman said on Brad's podcast [12:59] was Wait wait a minute. People got super [13:01] spooked and there was a lot of questions [13:03] about hey wait a minute. How are they [13:05] paying that 5-year $300 billion contract [13:07] to Oracle? Oracle stock fell 60% [13:11] and then time passed a little bit and [13:13] the models improved a lot and every [13:15] other day we're hearing that Claude is [13:17] at a new milestone in terms of the [13:18] revenue. So I think the market is [13:20] reacting to changing fundamentals. [13:22] >> Yeah, I think what happened was Claude [13:23] co-opted the market. Yeah. Right. I mean [13:25] people say there's two ChatGPT There's a [13:27] release in November 22 of the consumer [13:29] product and then there is Claude Code um [13:32] and the Opus model that came out [13:33] November-December last year and it just [13:35] kind of sparked a whole wave of [13:37] developer [13:37] >> Sonic 4.6. That's right. Yeah. [13:40] And yeah, I mean, just massive adoption [13:42] by by uh software developers and that's [13:44] just, you know, kind of completely [13:46] changed the narrative. The revenue [13:47] growth that um Anthropic announced was [13:51] like the most insane People's jaws [13:54] dropped. Cuz we all understand it's [13:55] growing, but I don't think people really [13:58] processed the sheer tens of billions of [14:02] dollars that are it's I don't want to [14:04] say out of thin air. No, it is. [14:06] >> It feels like it's I mean, it's not even [14:07] a public company. What So, listen. Do [14:09] the numbers. So, so, John, you have [14:11] chart four for a second. So, Oracle [14:13] stock is trading at a multi-month high [14:16] because there it is. Look at Look at [14:18] Oracle. Because, hey, wait a minute. The [14:20] numbers that we're hearing from uh from [14:22] these LLMs are actually pretty insane [14:24] and maybe the contracts are money good. [14:26] So, this morning [14:28] I listened to Patrick O'Shaughnessy [14:30] uh interview Krishna Rao, the CFO of [14:33] Anthropic. And Patrick tweeted a list of [14:36] surprising and mind-boggling stats in [14:38] this conversation. Here we go. Net [14:40] dollar retention is over 500% on an [14:42] annualized basis. Anthropic's first [14:45] >> What's net dollar retention? Basically, [14:46] how much money you are retaining and [14:48] growing from your customers. [14:50] >> This is for Anthropic. [14:51] >> So, 100% is like, all right, great. [14:53] Nobody canceled in the aggregate, but [14:55] you're not growing that that those [14:57] contracts. 500% on an annualized basis. [15:00] Anthropic's first dollar of revenue came [15:02] in March of 2023. March of 2023. [15:05] >> Yeah. Over 90% of code inside Anthropic [15:08] is written by Claude Code. The head of [15:10] tax is the heaviest token user on the [15:12] finance team. Run rate revenue went from [15:14] 9 billion to 30 billion in one quarter. [15:17] >> Sequentially. Quarter over quarter. [15:19] Reportedly on pace for 50 billion by the [15:22] end of next month. Cowork is growing [15:25] faster than Cloud Code did at the same [15:26] point in its life [15:28] and signed he said this on the podcast. [15:31] Signed two double digit million dollar [15:34] commits in a 20 minute Uber ride to the [15:37] podcast. [15:38] We've never seen this [ __ ] before. [15:40] What do you What do you think? Are we [15:41] like overly excited about this or is [15:44] this like [15:44] >> You can't possibly be overly excited. [15:46] It's insane. [15:47] I mean here's the thing though. If you [15:48] annualize at that rate for the next 2 [15:50] years [15:51] Anthropic is the biggest company in the [15:52] world. [15:53] >> You love the Fed's balance sheet. Yeah, [15:54] it's got to slow down. So it's got to [15:55] slow down. Um and and it's it's but it's [15:57] impressive. Yeah, I mean certainly [15:58] impressive. Kudos to them. Where is this [16:00] money coming from? It's got to be coming [16:03] It's got to be spending that would [16:04] otherwise have gone somewhere else. It's [16:06] It's not like new money has materialized [16:09] to contribute to a 50 billion dollar [16:10] revenue run rate for this company. [16:12] >> Is the answer everywhere? Yeah, [16:13] everywhere. I mean tech tech startups, [16:15] enterprise companies. [16:16] >> Industrials, everyone. [16:17] >> Other hedge funds, you know, [16:18] multi-million dollars. There's a lot of [16:19] big financial firms that are now totally [16:21] on board. Like I've talked to some huge [16:22] BlackRock and those kind of places are [16:24] all in on this stuff. BlackRock launched [16:26] I was at the New York Stock Exchange [16:27] today. This the Cerebrus IPO which we're [16:31] going to talk about in a second [16:32] was going public on the Nasdaq. [16:35] But on the New York I think BlackRock [16:36] had a Digital Realty Trust IPO. I think [16:40] that either they were celebrating it [16:42] today or it actually happened today. I [16:44] don't even I don't even know but like to [16:46] that point everybody is everybody is all [16:49] in on this and again it's the only [16:51] economic growth story that there is [16:53] right now. There's not a second one [16:55] unfortunately. Um Cerebrus raised five [16:59] and a half billion. [17:00] They sold 30 million shares. [17:02] Um plus four and a half million more [17:04] which I'm sure they'll do the green [17:05] shoe. Um [17:08] valuation out of the gates What what [17:10] what was [17:11] >> Wait, we don't know the We don't know [17:12] the close. [17:12] >> priced it at 185 I think, right? That's [17:15] the number? It opened at 350. It's It's [17:18] now 326. So, is this a $100 billion [17:20] market cap? I think that's what I think [17:21] that's what it would be. And they did [17:23] they did a I think it I think they did a [17:24] billion dollars or they're they're on a [17:26] billion dollars AR. Now, obviously we [17:28] can all do the math. That's absurd. [17:29] They're Why can't it trade 300 times [17:31] revenue? [17:31] >> But [17:32] Uh so, I I don't know if the I don't [17:34] know anything about this business. I'm [17:35] not going to comment on the valuation. [17:36] They're very very their revenue is [17:37] growing rapidly. Order book was [17:40] oversubscribed 20 20X. [17:43] They said at the open of trading they [17:45] had five orders uh five orders to buy [17:48] for every share available. Like [17:51] And [17:52] >> is that who are those orders coming [17:53] from? Retail. Um [17:56] Everyone is in already. This is so Every [17:59] every institution already owns it [18:01] because now we have IPOs before we have [18:03] IPOs. They did a series H round. You [18:06] know how much fun [18:07] Fidel [18:08] Fidelity is in the stock. How many [18:10] letters do we have? Fidelity is in the [18:12] stock in size prior to the IPO. So, [18:15] every everyone's already in it. So, it's [18:16] only retail. Um this is the biggest US [18:20] tech IPO since Snowflake. [18:23] They raised 3.8 billion in September of [18:25] 2020. It didn't go well after the first [18:28] day. The stock promptly collapsed, but [18:30] whatever. It's still, you know, the [18:31] company's fine. Um [18:34] Actually, hold on. Snowflake also went [18:36] public at 100 plus times revenue. Yeah. [18:39] So, the story here is they have a master [18:42] agreement with OpenAI for 750 megawatts [18:46] of inference capacity, which the only [18:48] person in the room that knows what that [18:49] is is Kai. Um [18:51] expandable to 2 gigawatts by 2030. [18:55] Um [18:56] They actually were trying to go public a [18:57] year and a half ago. They pulled the [18:58] S-1. [19:00] The concern was customer concentration. [19:03] People didn't want to buy a [19:03] semiconductor company that effectively [19:06] had like two or three really giant [19:07] companies. Nobody cares about that [19:09] anymore. So, that's the num biggest risk [19:12] in the S1 is that OpenAI contract and I [19:16] think they have a big deal with the UAE. [19:18] Risk. [19:19] >> So [19:20] yeah, I don't even know why people are [19:21] worried about that. What wait what what [19:22] does the company do? They're trying to [19:23] compete with Nvidia? [19:24] >> Inference thing. So they're not doing [19:25] tiny little [19:27] uh nano chips. They're doing big [ __ ] [19:29] wafers. This is a technical term. And [19:33] the idea is that the memory sits right [19:34] next to the compute so that there's no [19:36] latency. And this is what they say is [19:40] the ultimate chip for inferencing. So [19:43] ex- explain Can you explain it any [19:44] better or what What did I miss? I mean [19:47] look that that that's the high-level. I [19:48] mean I I think this is [19:50] This is kind of exactly what you'd [19:51] expect to happen in a time like today, [19:53] right? Nvidia has these huge profit [19:55] margins, growing like a weed at huge [19:57] scale. [19:58] What are you going to do? You're going [19:59] to compete with them, right? Their [20:00] margin is just so attractive. [20:01] Exactly. So whether it's this company or [20:03] or the next one or you know, even [20:06] Nvidia's customers, right? Google and [20:07] Amazon, everyone's trying to get in the [20:09] chip game because that's where the money [20:10] is. Um so this is, you know, very very [20:12] much in in line with where I think we [20:14] are in the cycle. So how do you how do [20:15] you talk about something like this [20:16] without sounding like a cheerleader? [20:17] Someone asked me yesterday like how [20:19] What's the difference between now and [20:21] the dot com bubble? And I said the [20:22] biggest difference now is probably that [20:23] the financial media was a bigger [20:24] cheerleader back then, but I don't know [20:26] how you as an analyst can talk about [20:28] what's going on and these growth rates [20:30] and what these companies are doing [20:31] without sounding like a permabull and a [20:33] cheerleader. It's impossible. [20:36] Well, the earnings growth is there. So [20:37] that's how you do it. [20:39] Um [20:40] Sean put this together for me for the [20:41] show today. [20:42] I'm not saying like the semis are cheap, [20:45] but they're 27 times forward earnings. [20:48] And that is inclusive of the fact that [20:50] the SMH is up 220% since liberation day. [20:54] So yeah, the stocks are up huge, but so [20:56] is the earnings growth. And a 27 [20:59] multiple does not rhyme with 1999. All [21:02] right. I I've I've I've another side the [21:04] other side of that. Not to not anything [21:06] you what you said is correct. So, [21:08] Nvidia, the stock has gone sideways from [21:13] August of 2025 [21:15] to today. So, that's like a that's a [21:17] long sideways digestion. Considering [21:19] that the market's been generally pretty [21:21] positive. Um and then of course it broke [21:23] out as we're speaking, had a massive [21:25] move over the last couple of weeks. [21:26] >> Put the chart put the chart up. [21:27] >> 7 days [21:29] Nvidia has added $900 in market cap [21:32] which is as large as McDonald's, Disney, [21:35] Boeing, Uber, [21:37] uh Starbucks and Royal Caribbean [21:39] combined. This is 7 trading days. [21:42] >> [laughter] [21:42] >> What are these numbers? These numbers [21:44] almost don't mean anything anymore. [21:45] >> All right. So, in terms of thinking [21:47] about Nvidia and [21:49] uh the valuation that it should trade at [21:52] Nvidia is too big to get a market [21:54] premium. Next chart, please. [21:56] It's one thing for Nvidia to trade at 50 [21:59] times forward earnings or 60 times [22:00] forward earnings when it's earning $20 [22:03] in in net income. [22:06] So, you could look at the chart on the [22:07] right and say say, "Oh, the forward P [22:09] has gone down. It's gotten cheaper." [22:10] It's like, "Dude, come on. Give me a [22:11] break." It's $223 [22:13] billion or what they're expected to earn [22:15] in net income over the next 12 months. [22:16] It can't possibly trade at a premium. It [22:18] is too large. [22:20] >> the valuation be? What would the [22:21] valuation be if it traded 30 times [22:23] earnings? [22:23] >> glad you asked, Josh. Look at the next [22:25] chart. So, chart goat, Matt [22:27] said [22:28] >> is working hard today. This is awesome. [22:30] >> So, check this out. So, right now [22:32] uh at a forward PE of 25 [22:35] Nvidia is at a $6 trillion market cap [22:38] give or take a little bit less. And it's [22:39] 8.5% of the index. The reason why I say [22:42] it can't possibly trade at such a large [22:43] premium is because it would swallow the [22:45] index. If it was still trading at 45 [22:48] times forward earnings, it would be $10 [22:51] trillion [22:52] and it would be 15% of the market. [22:54] It has to have a size discount. [22:56] >> Yeah, there'd be two companies, Nvidia [22:57] and Entropic. Right. [22:59] Right. So, I think I think the way that [23:01] the market has been treating Nvidia, [23:03] letting it digest, even though the [23:05] revenue and the net income and the [23:06] margins keep going up to the right, I [23:08] think it makes total sense. Perfect [23:09] sense. [23:10] >> week or 2 weeks ago it was like Google [23:12] almost reached Nvidia's market cap. Now [23:14] Nvidia has a trillion-dollar lead on [23:16] them. Cuz there's enough shares [23:19] and shares outstanding plus the gain. [23:21] Jesus. [23:21] >> You know what's so crazy though? [23:23] The stock, to Michael's earlier point, [23:25] sat at 180 for 6 months. [23:27] Like any You could have bought as much [23:29] as you Now, where is it now? 230? So, [23:32] you were you were saying this to Brett. [23:33] I thought you you made a really good [23:34] point. [23:36] It was So, it was 235. It was there for [23:38] almost a year. You could have picked it [23:40] up. Under two under 200 you could have [23:43] bought as much as you want. [23:44] >> But you know what? I understand why [23:45] investors didn't because in October when [23:47] we were in Austin, it reported a monster [23:49] beat. A monster monster monster beat, [23:51] like an LOL type beat. The stock popped [23:53] up 4% and it ended up closing down 4% [23:56] and it was very understandable for [23:58] investors to say, "All right, I guess [23:59] the trade's over." [24:01] And it was until it wasn't. [24:03] Is it irrational to have Nvidia be a 10% [24:06] position in a in a [24:08] uh in a portfolio? [24:10] Seems like a big position. [24:12] >> [laughter] [24:13] >> But is it irrational given the size and [24:15] scale of the business and how important [24:17] it is to everything else? [24:18] >> Kai, last week Adam Parker was saying [24:20] Nvidia is a sector. So, back to Josh's [24:22] question. If you view it through that [24:23] prism [24:24] >> of accelerated computing is a sector, [24:25] but they have competition, right? As we [24:27] discussed, Cerebras just IPO'd today to [24:29] to compete against them. We have their [24:31] biggest customers, Google and Amazon, [24:32] also competing against them. So, like it [24:34] it, you know, the the history of this [24:36] stuff is is is cyclical, right? Nvidia [24:38] was one of the most cyclical, you know, [24:40] companies for a long time until until [24:41] more recently. So, I think you have to [24:43] take that into account. I mean, this [24:44] goes to the the question of like, yeah, [24:47] obviously the market's up big and it's [24:49] all it's up big on the back of [24:51] fundamentals, earnings that have [24:52] increased. The question is not so much, [24:54] you know, are valuations extended [24:56] relative prices extended relative to [24:58] earnings. It's more how sustainable are [25:00] these earnings to the extent that [25:01] they're all kind of downstream of one [25:04] one phenomenon, which is massive capex [25:06] by the hyperscalers into building out a [25:07] data centers, which goes feeds the chip [25:09] companies and feeds the feeds the the [25:11] power companies. It's all one trade. [25:13] Yeah. And so that's the that's the big [25:14] question, right? Which is like at the [25:16] end of the day what what matters is will [25:18] the enterprise adopt AI? Will, you know, [25:21] right now all the all these CEOs Yeah, [25:22] the answer is yes. I think the answer is [25:24] yes. No, you know the answer is yes. [25:26] >> We know the answer is yes. [25:27] Well, otherwise where does $30 in [25:30] quarterly revenue [25:32] in revenue run rate for Anthropic come [25:34] from? That's not Well, well, well, well, [25:35] it comes from, you know, folks following [25:37] in, right? Some from like business [25:39] leaders being told, "Hey, by their [25:40] board, hey, if you don't adopt AI, you [25:42] don't digitally transform, then you're [25:43] fired." Yeah. So, I think right now [25:44] there's, you know, obviously an arms [25:46] race by corporate America to say, "Hey, [25:48] we're on top of the ball. We're doing [25:50] stuff." Now, the question is if they [25:52] will they actually generate meaningful, [25:54] um, like revenue boosts from this [25:56] adoption. If so, then they'll keep [25:57] paying, and if not, they'll pair back [26:00] and say, "Hey, that was, you know, that [26:01] was an interesting experiment. We're [26:02] onto the next thing." [26:03] >> We had the company that's built our data [26:05] lake here yesterday. Um, they're called [26:07] Invent. They're geniuses. [26:10] And, uh, I wasn't like there for the [26:12] whole meeting. I, you know me, I just [26:13] pop in and say say the most outrageous [26:16] thing. [26:17] So, I walk in and, uh, [26:19] and he's sitting there with my [26:20] president, with like, you know, my vice [26:22] president, like going through all [26:24] And I'm just like, "Guys, let's cut to [26:26] the chase. Do we do an AI?" [26:29] >> [laughter] [26:29] >> They're like They're like, "Yeah, we're [26:31] doing an AI." I said, "All right, [26:32] awesome." You know, like that was my [26:34] contribution, but but it's sort of a [26:36] joke, but it's sort of not. Like every [26:38] business leader in every segment of the [26:41] economy at every company [26:43] just make sure we're doing some AI [ __ ] [26:45] Like that's what we have to do. Everyone [26:47] else is doing it. Not just do it, but [26:49] like let's do it in a way where I can go [26:51] back to my board of directors or I can [26:54] go to my shareholders and say we made X [26:57] dollars because we're doing this or we [26:59] saved X dollars because we're doing [27:01] that. [27:02] That's the pressure that every leader [27:04] feels right now, which is why none of [27:06] this seems like um optional spending. [27:09] Right. It seems like they're all [27:10] compelled. [27:10] >> I think like people talk about this AI [27:12] bottleneck and this kind of [27:14] fact that, you know, Anthropic's compute [27:15] constrained, but I think that there's [27:17] this, as you point out, [27:19] kind of FOMO, you know, in corporate and [27:22] then you couple that with this idea [27:23] that, you know, these businesses, [27:25] Anthropic and Open AI, are in a [27:27] competition with each other as well. [27:28] They try to try to lock the market up, [27:30] right? This is the kind of classic Uber [27:32] playbook. We want to be We think it's [27:33] winner takes all, winner takes most. We [27:35] want to win market share. And so [27:36] therefore, they subsidize token costs. [27:39] Like token costs are below what it below [27:41] what they should be. These companies are [27:42] not profitable. Yet they're willing to [27:44] run at a loss because they want to [27:45] capture market share. Right? So it's not [27:47] a true price signal. The fact that we [27:49] are, you know, compute constrained now [27:51] doesn't really mean anything. So I think [27:52] like we should kind of look past that [27:54] and ask the question of do we actually [27:55] think the technology will be, um, you [27:58] know, useful for driving ROI for [27:59] businesses? And if the answer is yes, [28:01] you know, how much can the labs and the [28:03] hyperscalers capture of that versus [28:06] their customers? [28:06] >> What do you think? [28:09] I think the answer is yes. I think that, [28:10] you know, having, you know, I was an [28:11] early adopter of a lot of these tools. [28:13] I've been, you know, I trained my own [28:14] LLMs starting in 2019, 2020. I think [28:16] this stuff is for real. Um, I think the [28:18] technology is tremendously useful. Um, [28:20] obviously it's not perfect. You know, [28:21] certain things I wouldn't trust AI for, [28:23] but in terms of like, you know, kind of [28:25] lower level stuff, you know, [28:27] basically the history of disruption is [28:28] this, that whenever a new technology [28:29] comes out, you want to start by giving [28:31] it kind of low end tasks where mistakes [28:33] are forgivable. And I think, you know, [28:36] at first AI was a good analyst. Now, you [28:38] know, I'll give you an example as a [28:39] quant. So I'm a quant. I, you know, do I [28:42] I code a lot. I run models. [28:44] >> accept you in this room. [28:44] >> You can, yeah, you can tell I'm a quant, [28:46] right? I mean, I threw a blazer on to [28:47] try to disguise it, but No, we know. Um [28:49] he's like a he's like a like the bun [28:51] with a quant. That's a That's a That's [28:52] That's a first, right? [28:53] >> quant. Okay. [28:54] >> [laughter] [28:54] >> There you go. Yeah, I see. Yeah. [28:57] No, [clears throat] so so I used to you [28:58] know, I had I used to employ analysts [28:59] whose job was to you know, I'd give them [29:02] as a research director tasks. Hey, I [29:03] want you to go you know, run this [29:04] experiment. See how this factor would [29:06] have performed in this market. Um you [29:08] know, Cloud Code is or Codex as well are [29:10] you know, perfectly capable of doing you [29:12] know, a lot of that analysis. Hey Codex, [29:14] here's the API key, you know, go to this [29:16] database. Here's a schema. Here's a [29:17] script. Can you mimic this? I want you [29:19] to study X. And it'll come back and and [29:21] you know, we can iterate together. So I [29:23] think like it's starting to climb up the [29:24] um the the ladder. [29:26] >> the place of you having a back and forth [29:28] with another quant [29:29] who you're assigning things to. [29:31] Now you're talking directly like the [29:33] user interface is like [29:35] it's it's it's call and response. That's [29:37] right. [29:38] >> You're saying things to it, it's saying [29:39] things back. [29:41] That prompts you to say the next thing. [29:42] It's conversational. Yeah, it's an [29:44] iterative of uh [29:46] uh Right. process. [29:47] >> the question I have for you and [29:49] I have like the cognitive dissonance in [29:51] my brain has been firing for months. [29:52] Like I know the history of this stuff, [29:55] but I also know that sometimes these [29:56] things are different. So like your [29:58] background fascinates me cuz you worked [29:59] at GMO with Grantham and Chancellor. And [30:01] so I I listened to your recent So you're [30:03] you got a great new podcast called the [30:04] Intangible Economy. You talked to Edward [30:06] Chancellor who's the author of one of my [30:07] favorite history books of all time, [30:09] Devil Take the Hindmost. I love that [30:10] book. I recommend it all the time. [30:12] And you were talking to him about [30:13] capital cycles. [30:14] And you're a very forward-looking [30:15] person. You're also a quant who knows [30:17] market history. [30:18] How do you deal with this [30:20] understanding history that hey, and [30:22] Chancellor was basically saying listen, [30:23] every time this happens, we spend too [30:25] much money, these capex cycles, they [30:27] follow a pattern. This is going to end [30:29] in tears. [30:30] Versus [30:32] I think Templeton one time said listen, [30:34] 20% of the time it really is different [30:35] this time. So how do you how do you have [30:37] cuz I have those competing thoughts in [30:39] my brain all the time? Yeah. That [30:41] nagging suspicion that this is going to [30:42] look like every other CapEx boom that's [30:45] ever happened where they'll go way [30:48] overboard with spending and we're all [30:49] going to pay the price for it in the [30:51] form of stock prices. [30:52] >> How how do you handle that? [30:54] Well, so I think [30:55] >> Make us feel better about it. What's [30:56] your secret, Kai? [30:57] Look, I think I think two things can be [30:59] true at once. I think, you know, AI [31:00] technology can be transformative and it [31:02] may also be a bad investment. [31:04] The question is as as investors for us, [31:06] right? It's it's less, you know, it's [31:08] less will AI is AI a bubble or not. It's [31:10] it's that's too simplistic. It should be [31:12] where will the value accrue through the [31:14] value chain. [31:15] Will it will it be the model providers? [31:16] Will it be the chip makers? Will it be [31:18] the users? [31:20] You know, Chance was framing was really [31:22] interesting. So he's he's a capital [31:23] cycle theorist. He studied, you know, [31:25] the booms and busts around, you know, [31:27] electricity, the railroad, the canals, [31:29] the dot-com boom, you know, and and in [31:31] every case, aside from one, telephone [31:34] being the one example, one exception [31:36] because it consolidated into monopoly, [31:38] but in every single other case, what [31:39] what's happened is all the capital comes [31:41] into the sector on the supply side to [31:43] build out, you know, the infrastructure [31:45] needed to run the new technology. But, [31:47] you know, they kind of get over their [31:48] skis, too much money comes in, demand, [31:50] you know, may materialize but perhaps [31:51] too slowly, so there's an air pocket. Um [31:54] and you know, and then the there's a [31:56] shakeout where prices fall, capacity is [31:58] unused, and the the the ironically, the [32:02] the guys who invented and built out the [32:03] infrastructure end up going bust, right? [32:05] Hundreds of railroads, hundreds of auto [32:06] companies, all the telecoms. What would [32:08] what would you say to this? That [32:10] companies know so much more about how to [32:12] run a business than they did 100 years [32:14] ago. We have such [32:16] foresight into how much what the deploy [32:19] supply demand imbalance might look like. [32:22] I it seems hard to believe that one day [32:23] we're going to wake up and be like, [32:24] "Whoops." I think I think the problem is [32:26] it's behavioral. I think it's companies [32:28] can't help themselves. You have like [32:29] Zuckerberg, you have, you know, the CEOs [32:31] of these companies saying, "I'd rather [32:32] go bankrupt than lose this race." Like [32:34] it's they they I think they all know. [32:36] It's like pure pressure. They have to, [32:37] right? Cuz you know, if you're running [32:39] company X and you know, you're like, [32:41] well well actually if [32:44] Sam Altman and OpenAI get AGI and I'm [32:46] not even doing anything, I'm done, [32:48] right? Like that that's the end of my [32:49] business. So I have to compete. And so [32:51] once Sam goes all in, everyone else has [32:52] to go in all into it. It's this game [32:54] theory, this the prisoner's dilemma. [32:55] >> the Is he the horse? [32:56] >> He's the ringleader, yeah. It's It's [32:58] still him. I think so. Okay. I heard [33:00] Bezos make some sort of a comment where [33:03] he thinks AI will be transformative and [33:05] beneficial for society, but [33:08] he also sees it as an industrial bubble [33:10] and most of the spent I I forget if he [33:13] said most or a lot of the money being [33:16] spent will end up being wasted. How [33:17] about this? What would surprise you [33:18] more? [33:19] This is a huge bubble that pops and AI [33:21] is still successful like the dot-com [33:23] bubble. Everything we wanted out of the [33:24] technology dot-com bubble, everything [33:26] they wanted and more happened, but we [33:27] had to live through the dot-com bubble [33:28] to get there. That's option one. Option [33:30] two is no, this is a perfect handoff. [33:33] The baton goes from spending to ROI and [33:36] uses and we're off to the races. Like [33:38] which one would surprise you more out of [33:39] those two options? I I just think so [33:40] many things have to happen, right, for [33:41] the baton to be handed off perfectly. [33:43] What a hater. [33:44] Yeah, I mean Why are you so bearish? [33:47] It's [33:47] At this scale, it's never happened [33:49] before, ever. We've never managed to [33:51] perfectly execute the handoff. I also [33:53] think that this cycle, this 15 whatever [33:55] year cycle, has had so many things that [33:57] have never happened before that I'm [33:58] willing to at least have an open mind [33:59] that it could. Yeah. Right? [34:01] >> Look, I I I think there's a chance that [34:03] this all works out. Um I think the the [34:06] downside risk um is probably not [34:07] favorable um if you're an investor in [34:09] the infrastructure side. [34:10] >> All right, you want to get nuts? Let's [34:12] get nuts. [34:13] >> [laughter] [34:14] >> Josh Chart 6. [34:15] All right, Jason Gabler tweeted, "Look, [34:17] I know none of this stuff matters [34:18] anymore, but my god, this will be the [34:20] fourth time the S&P 500 has hit a record [34:23] high while 5% of its members fall to [34:25] 52-week lows. Here's Here's the other uh [34:27] three dates. [34:28] July 1929, [34:30] not great. [34:31] >> [laughter] [34:31] >> January 1973 I December 1999 [34:36] Yeah, so we have a lot of stocks on the [34:38] 52-week low list. [34:40] Um for a market that's for a Dow 50,000 [34:43] party. [34:45] And uh you know [34:46] without even looking you already know [34:48] it's anything housing related. [34:49] And then it's a lot of software [ __ ] [34:51] And uh Equal weight up 7% [34:54] >> and a lot of consumer [34:55] >> Restaurants [34:55] >> of consumer stuff. Home builders, Home [34:58] Depot [34:59] Is that is that The K-shaped economy. [35:01] K-shaped economy [35:02] >> on steroids now. Does that bother Does [35:04] that bother you about the tech rally? [35:06] That it's not as broad as you'd like to [35:08] see it or do you not get worked up about [35:10] this stuff? [35:10] >> it's kind of two sides of the same coin, [35:12] right? Sammy Israeli and software sells [35:14] off. Right? It's it's almost the same [35:15] trade these days. I don't think it [35:17] should be. We can get into why that is [35:18] the case, but you know, that that is a [35:20] narrative today that the more powerful [35:22] is is AI, the more disrupted is [35:24] software. Can I tell you something? [35:26] Yesterday I was looking at software [35:28] charts. [35:29] I think they're going to zero. [35:31] I really do. I know they won't all. [35:33] Um but I am starting to believe that [35:36] they are newspapers. Hold on. That's [35:38] this is an absurd claim. So when you say [35:39] they're going to zero, be specific. [35:41] >> Like a lot of publicly traded software [35:43] companies [35:44] uh are going to be zeros. [35:46] Like not like zero zero, but net never [35:51] never coming never coming back. And [35:54] I I'm hoping it doesn't happen to the [35:55] big ones, but um I looked at like 20 [35:59] software company charts. Every single [36:01] one of them is either at a 52-week low [36:04] or close. [36:05] There aren't even like up weeks. [36:08] It Some of these charts it's like 25 [36:11] straight weeks lower close on Friday [36:13] than the open Monday. Yeah. It's [36:16] It's It's actually unbelievable. There [36:18] have also been zero takeovers. So can I [36:21] does a company like Adobe, let's use [36:22] them as an avatar. Do those companies [36:24] make it? [36:27] I think I think so. Um I think so. I [36:28] mean I don't know about Adobe in in [36:29] particular. I mean here here's the [36:31] question, right? Which is So, what why [36:32] are these companies down so big? They're [36:34] down so big because people assume that [36:36] their moat is code. And with code accent [36:39] cloud code, code's basically free, [36:41] right? Anyone of us here can can [36:42] effectively write code and and replace [36:44] Adobe Salesforce these products on our [36:46] own. Okay, so if if they have no moat, [36:48] then they actually go down. I guess the [36:50] question I would ask is whether code was [36:53] ever the moat for these companies, [36:54] right? Like I You think even You think [36:55] distribution was the moat? Yeah, I would [36:57] argue distribution, [36:58] >> can't So, right, but so then why can't [37:00] OpenAI and Anthropic mimic that [37:03] distribution? They certainly have the [37:04] capital. [37:04] >> They're trying to, right? So, that we [37:05] saw the open eye open AI development [37:07] company that this announcement this week [37:09] they're partnering with a bunch of PE [37:10] firms to essentially launch like a army [37:13] of four deployed engineers, basically [37:14] consultants, to go into your company and [37:16] >> recommend you use the the product. [37:19] >> our product and here's how you use it, [37:20] etc. Like they're certainly trying and I [37:22] think it's interesting because, you [37:24] know, I had this view for a long time [37:25] that the the luminaries on AI were kind [37:27] of like these naive people that they [37:30] kind of thought if we build it, they [37:31] will come. Let's just build a really [37:32] cool product and then the enterprise [37:34] will will buy it. I think what they're [37:35] what they they've learned correctly is [37:37] that you have to sell, too. You have to [37:39] push this into the enterprise. Like [37:41] businesses are very slow to move, [37:42] they're very slow to change. Um I think [37:44] this is [37:45] an important recognition. They've also [37:46] done partnerships with, you know, the [37:48] the old school consulting firms, the [37:50] Accentures of the world. Um and then you [37:52] we saw Google is hiring a bunch like a [37:53] thousand or something. [37:55] >> Yeah, you have to get to the decision [37:56] maker and you have to push them over. [37:57] But the market seems so convinced of it [37:59] that these companies are, to Josh's [38:01] points, uh metaphorical zeros. Like [38:03] Adobe is the the the earnings per share [38:05] and the free cash flow per share are [38:06] still at all-time highs. I'm guessing [38:07] that Salesforce is not has not seen any [38:09] material contraction. [38:11] But the market is like, "All right, we [38:13] don't care. You're down 60% anyway." [38:14] Yeah, I I think Adobe's trading at a PE [38:16] of like 9 and 1/2. [38:17] >> [laughter] [38:17] >> it was at 10 yesterday. Yeah, and like [38:18] Salesforce is trading [38:18] >> auto That's an auto maker PE. What we're [38:21] saying is that [38:22] what we're saying is that those earnings [38:24] are not going to show up next year. [38:25] That's the only answer. [38:26] >> A software company with high margins [38:28] trading below 10. So this the the stock [38:30] market is like the economy in that we've [38:31] had these rolling recessions in parts of [38:32] the economy, but it hasn't brought the [38:34] whole economy down. And the stock market [38:36] is the same thing where we're having [38:37] these these losers get separated. So the [38:39] question is, could we have this whole [38:41] thing with AI without bringing the whole [38:42] market down? And Well, the semis the [38:44] semis are gaining more and market cap [38:46] than the software companies are losing. [38:48] So it's a net positive for the Nasdaq. [38:50] It's a net positive for the S&P. To to [38:52] that to that question, like yeah, we're [38:54] sort of seeing 25 fairly large companies [38:59] disappear before our eyes, [39:02] but then there's 25 other very large [39:04] companies that have just become [39:05] gigantic. Like top 20 S&P market caps. [39:08] >> Yeah, this chart here shows that the the [39:10] SOX index, the semiconductor index, [39:12] these companies are now 23% of the S&P. [39:15] Right. Out of nowhere. [39:16] >> It's a baton it's a baton being handed [39:18] off. 20 [39:19] 23. [39:21] Yeah. [39:21] >> 23 coming up. [39:22] So the market so the overall market is [39:25] not suffering, but there are gigantic [39:28] companies whose market cap is literally [39:30] vanishing. [39:31] I So I say like zeros, I don't mean like [39:33] there's no more earnings and revenue. I [39:35] just mean like these stocks that might [39:37] never come back again. [39:39] They could just And I those are take [39:41] private at some point. But that's that's [39:43] what I'm asking. Where are the deals? [39:45] Nobody thinks these are cheap enough to [39:47] LBO. It's way too quick for deals, don't [39:48] you think? At this point it's [39:50] >> So But if we think there's $3 trillion [39:52] in dry powder [39:53] um amongst private equity, private debt, [39:56] whatever. Like we know there's a lot of [39:58] money they want to put to work. [40:00] Take one. But they're [40:01] they're already over subscribed in in in [40:03] software, right? They're they're they're [40:04] full. Right. So [40:05] >> [laughter] [40:06] >> I have private credit fund and also [40:08] we're buying Adobe. Yeah. Yeah, why not? [40:11] Can you imagine? Imagine the media [40:13] coverage of that. So wait, so you you [40:15] mentioned like the earning side of this [40:16] thing. And and we've got a million [40:18] charts in here that shows how great [40:19] earnings are doing. Michael, why don't [40:21] you like pull up some of the Nasdaq ones [40:22] that we Michael and I talked to a guy [40:23] from Nasdaq for Animal Spirits the other [40:25] day. All right, let's let's do chart 14. [40:27] And like when you look at the earnings, [40:29] this is the thing that like I said this [40:31] is the most logical melt-up of all time [40:32] cuz it really doesn't look as much like [40:34] '99 when you think about like the way [40:36] that these companies are. So in 1999 in [40:39] in in in the in the fourth So at the top [40:41] at the top of the bubble, [40:43] um [40:43] 10 and 1/2% of the Nasdaq 100 had [40:46] negative margins. [40:48] There was no companies that had margins [40:51] between 50 and 100%. [40:53] Uh [40:54] and it's just completely lopsided. So [40:55] today today 20% of the Nasdaq 100 has [40:59] margins between 50% and 100%. [41:01] 50% of the index is between 25 and 50. [41:04] That compares with just 24% from 1999. [41:08] Back then most of the companies had a [41:09] margin between 10 and 25% or 0 to 10%. [41:14] These were not the same businesses. Now [41:16] Dan Greenhouse made a fair point. Dan [41:18] basically said [41:20] the biggest myth is that the tech [41:22] companies today are real and in the '90s [41:25] these were like just fugazi companies. [41:27] And he said that's not true. [41:29] We had Cisco, [41:30] uh Microsoft, Oracle and at the time [41:33] Cisco was doing $19 billion in revenue. [41:35] Microsoft was 22. I think if you [41:37] inflation adjust this, so Microsoft was [41:39] doing 23. Inflation adjusted that's 44 [41:41] billion. Cisco was 19. [41:44] $38 billion inflation adjusted. So Cisco [41:46] was doing what OpenAI is doing. Like [41:48] these were these were massive companies. [41:50] It's just that a lot of the rest of the [41:52] index was pure [ __ ] Well, that's the [41:54] thing. We had pre- forget about profit [41:56] margins. We had pre-revenue companies. [41:59] We had thousands of IPOs that we don't [42:02] have right now. [42:03] Like that's not what's happening. John, [42:04] give me 16.2. So 16.2? [42:07] >> Holy [ __ ] Holy [ __ ] [42:10] Are you sure you want to whip this out? [42:12] >> [laughter] [42:13] >> Turkey did this for me. So, this is [42:14] NASDAQ 100 earnings growth. For 20 [42:16] years, this thing has printed 14% annual [42:20] earnings growth. So, you had Mobius and [42:22] on your podcast. He always talks about [42:23] baselines. [42:24] I can't back this up. I I'm just [42:26] guessing. There's no way we've ever had [42:27] a period like this with earnings growth [42:29] this high for this long before. So, in [42:31] terms of baselines, the question is [42:33] could this continue? Could we see [42:35] earnings So, yes, cuz the [42:37] the NASDAQ is up 20% per year something [42:40] over this over this period. [42:43] Is is AI again back to the baton handoff [42:45] thing. Can AI keep this kind of earnings [42:47] growth going? [42:48] >> Uh robotics could do it. [42:51] You want another decade You want another [42:52] decade of 15% revenue growth? [42:55] How about uh 10 million robots? Hu- How [42:58] about a million humanoid robots and [43:01] automating literally every inanimate [43:04] object in the in the entire world? Cuz [43:06] that's That's what's coming now. Golden [43:08] era of biotech. [43:09] Yeah, well, that could do it, too. We'll [43:11] live live to 120. Yeah, that'll that'll [43:14] move the needle. So, that's how you get [43:16] 50 That's I think uh [43:18] I was talking to my friends who are not [43:19] professional investors. Um [43:22] but [43:23] I said the word maybe the worst thing [43:25] you could do is have too much cash at [43:28] too young of an age. [43:30] Not that the market won't at some point [43:31] have a horrific event where you wish you [43:33] had more cash. But like that's going to [43:35] get bought really fast all over again. [43:38] Which part? [43:40] Uh if we have another if we have like [43:42] anything non-recession, any kind of [43:44] market crash, right? That'll get bought [43:47] up in 2 seconds. We'll get another V. We [43:49] had two. Yeah. We'll get another V. You [43:51] know why? [43:52] Cuz people see the degree of change [43:54] happening and the speed and the [43:57] acceleration of automation. [43:59] And I think they realize they can't not [44:01] own the chips, the robots, the AI. [44:04] So, I'm not saying like the market's not [44:06] going to fall. It definitely will. [44:08] I'm just not convinced we're going to [44:09] have like the a 2-year bear market with [44:12] this much innovation happening. [44:14] >> I remember back in March of Oh, this [44:17] year. This was 2 months ago. [44:20] Micron was at 470 [44:22] had an another earnings report that we [44:24] all laughed at. The numbers were like [44:26] Wait, what? And then Micron swiftly went [44:29] from 470 down to 310. [44:33] In March in like [44:35] 10 sessions, it fell 30%. [44:38] It went from 310 to 800. Yeah. In a [44:42] month or two. [44:43] >> Right. If this If this cycle turns for [44:45] whatever reason, it's going to fast. [44:46] John, give me chart 16 real quick. 0.3 [44:49] This is the melt-up. Matt did this for [44:51] me. So the Nasdaq in the '90s, like [44:52] we're the Nasdaq 100 was the best 10 [44:54] years. It's kind of approaching that. [44:55] But after the '90s, the Nasdaq fell 83% [44:59] and it took like 12 years to break even. [45:01] That's the kind of thing we're not going [45:02] to get again. I don't think so. [45:04] >> I don't think so. There's no way you can [45:05] I don't think we can have that kind of [45:06] fall for that long again. I agree with [45:07] that. [45:08] >> Chart 15.9 [45:10] So [45:12] >> [snorts] [45:12] >> in the dot-com bubble [45:14] 32% of the index constituents were [45:17] trading between 60 and 100 times [45:20] earnings. [45:21] 34% of the index was trading over 100 [45:24] times earnings and 10% of it was [45:25] unprofitable. Today, 60% of the index is [45:29] trading between 20 and 40 times. Between [45:31] 20 and 40. And 20 is trading between 40 [45:33] and 60. We have 10% of the index that's [45:36] trading at stupid levels. But we're not [45:38] We're just not valued the way that we [45:39] were in the 1999. So yeah, the market [45:41] could cut in half. It always can. But is [45:43] it going to fall 80%? That That would be [45:45] very hard to believe at this point. [45:47] What do you What do What are you worried [45:48] about in this in the midst of like [45:50] everything that's going on? What do you [45:52] What do you think is the big risk? Maybe [45:54] even if it's an obvious risk, what do [45:55] you think is the thing that we should be [45:57] focused on? [45:58] I still think that AI is a risk if only [46:01] because it's the entire market, right? [46:04] Like the S&P's, you know, 33% mag seven, [46:07] you add in these chip stocks, you're [46:08] talking 50, 60% of the passive index [46:11] which we all told that that's like [46:12] diversified is one theme. And so like so [46:15] like yes, I mean as things you could [46:17] things continue to be going well as they [46:18] have today and the past month, that's [46:20] that's fantastic, but I just think [46:22] there's a lot of concentration. [46:23] >> such a big part of the economy now that [46:25] if this turns Michael and I were trying [46:27] to figure out what that mean, like what [46:28] would how far would they have to pull [46:29] back on the spending, but you'd you'd [46:31] probably get a recession and a bear [46:33] market together. It wouldn't just be [46:34] one. There's a there's a different [46:36] answer to that though. We all assume [46:38] that this ends badly [46:40] because of the hiccup in the capex story [46:42] and people have been saying that now for [46:43] three years. [46:44] >> Yeah. I've said it. All right. So [46:46] everyone [46:47] everyone but like the other risk is that [46:49] this works really really well [46:52] and and the job displacement just hits [46:54] harder and faster than any of us expect. [46:57] And that becomes an economic risk [47:00] which [47:01] I don't know if that affects the Nasdaq. [47:03] I don't even know if the Nasdaq and the [47:05] actual economy have any relationship [47:07] whatsoever left. But to me [47:09] >> if Anthropic and OpenAI are in the [47:11] Nasdaq at that point. Yeah. [47:13] >> I had a family member who said, "I'm [47:14] worried about my job because of AI." You [47:15] know what I said? Buy stocks. Well, I've [47:18] been saying that for 10 years. Just [47:19] don't let the damn robots. I mean, I [47:21] don't know I don't know what your [47:22] alternative but like do you do you think [47:24] that that is a [47:25] underappreciated risk that the job loss [47:28] like materializes and it's bigger than [47:31] people thought? [47:32] I think it's a possible risk but but I I [47:34] actually think that that's an overrated [47:36] risk. I think, you know, the average [47:37] person, you know, because they've been [47:39] watching sci-fi movies, um you know, [47:41] over indexes on that as a possible [47:43] future and relative to what I think is [47:45] actually attainable. Now, never say [47:46] never, things can happen that um you [47:48] know, we we can't predict. But I'd say [47:50] that that is, you know, probably not not [47:52] the first thing I'd be worried about. Um [47:54] electricity. You worried about that? [47:57] Not having enough of it? [47:58] >> Yeah. [48:00] In the short term [48:00] >> Could that stop this CapEx boom from [48:02] booming? Would that be the the [48:04] fundamental constraint? So, we've heard [48:05] about chip shortages. We understand we [48:07] short compute. Could the Could the [48:10] strain on the grid be the thing that um [48:15] diminishes our chance of hitting these [48:17] earnings expectations? [48:19] Yeah, look, there there are a lot of [48:20] bottlenecks potentially in the value [48:21] chain. And you know, it's weird cuz if [48:23] you step back, I almost feel like it's a [48:24] good thing, right? So, like talking [48:26] about a job loss scenario. [48:28] >> it's like a governor. [48:28] >> It's a governor. It kind of slows down [48:30] the build. Like if things happen too [48:32] fast, the government, you know, [48:33] regulations are too slow to adapt. The [48:35] job market, you know, re-trading [48:36] employees and restructuring companies [48:38] happens too slowly. So, in a way, the [48:39] best case scenario is one in which AI [48:41] ends up delivering abundance, but it [48:43] takes like many years for that to [48:45] happen. The scenario that we worry about [48:46] is one in which that happens overnight, [48:48] and then everyone's laid off, and it [48:50] creates like a doom loop. [48:51] >> So, how about like getting [48:52] everyone at risk? So, you got a chart in [48:54] here, John Deere 2028. So, you have to [48:56] talk about how the Mag 7 is going from [48:58] intangible asset light to asset heavy. [49:01] What if it's just a re-rating? And they [49:03] go, all this spending and you becoming [49:05] more of data centers and and the [49:08] physical. It's not intangible anymore. [49:10] What if we just get a re-rating that way [49:11] of valuations? That is my base case, [49:13] actually. So, I I don't I don't think [49:15] these companies are going to go bust. [49:16] >> So, a re-rating lower? Yeah, if we get a [49:18] re-rate just valuations have to be lower [49:20] cuz you're more capital intensive. [49:21] >> So, like like let's use an let's use an [49:23] example. Let's take an Alphabet. We love [49:25] this business for so long because of how [49:27] uncapital intensive it was. [49:29] Base- right? Basically, they invented [49:31] search 25 years ago, and they've been [49:33] eating They've been dining out on that [49:35] innovation. And the margins were crazy. [49:39] And then cloud computing, amazing [49:41] margins. [49:42] Now, obviously, nobody's calling AWS and [49:46] Alphabet overall like or or or or [49:49] Microsoft. Nobody's calling these [49:50] companies [49:52] um cap light. They're the opposite. [49:54] >> So because of your your research on [49:55] intangibles, you're saying that's your [49:56] baseline now. John, throw that chart [49:58] back on. Kai, speak to speak to what [49:59] we're looking at here, please. [50:01] So we're looking at the Mag 7 CapEx to [50:03] rev CapEx as a percentage of revenue. [50:05] CapEx to sales ratio. [50:06] >> Okay. [50:06] >> Yeah, so This one. Yeah, to Josh's [50:08] point, like the the Mag 7, why are they [50:11] the Mag 7? Why are they why do they have [50:12] their own acronym? It's because they [50:13] just have delivered amazing ROIC, return [50:16] on invested capital, over the past, you [50:18] know, 20 years, 15, 20 years. [50:20] Unprecedented. [50:21] >> Unprecedented. Never seen before. Never [50:22] seen before at such scale. Right? Like [50:24] yeah, okay, it's possible to build [50:25] asset-light businesses that are small, [50:27] but at the trillion-dollar scale, I [50:28] mean, no one thought it was possible. [50:30] And and the and the way they've been [50:31] able to do that is through these [50:32] intangible assets, through leveraging [50:34] brand, human capital, [50:35] >> Network effects. [50:36] >> Network effects in particular for [50:37] Google. Um and and that was a great [50:40] thing while it lasted. And so here's the [50:41] irony is that these guys are, you know, [50:44] on the forefront, they invented Google [50:45] invented the transformer. They're now [50:47] kind of in a way engineering their own [50:48] demise because they invented the [50:50] technology, what does it do? Well, it [50:52] changes the the rules of the game. [50:54] Entering the AI revolution, the way it [50:56] worked was there was a handful of [50:57] digital services and these businesses [50:59] kind of like carved it up into their own [51:01] fiefdoms. All right, you get search, you [51:02] get social, you get shopping, and it was [51:04] a good cozy little oligopoly to have. [51:07] What happened is with AI, now, you know, [51:09] the the perception at least is that it [51:11] collapses all the markets into one. That [51:13] now it's all about who gets agents first [51:14] and who wins that market. If you win AI, [51:16] you win everything. And that's ignited [51:18] this game theoretical prisoner's dilemma [51:20] situation where you have all these [51:21] companies saying, "Wait, this is an [51:22] existential risk. I need to They're [51:24] still spending therefore on If they're [51:25] spending, I got to spend. Right? [51:26] Ideally, they would all moderate their [51:28] investment, be incremental, continue to, [51:30] you know, make AI not a disruptive but a [51:32] sustaining innovation um for the [51:34] incumbents. In this case though, [51:35] however, if Altman's going to spend the [51:37] trillion dollars, then you got to spend [51:38] the trillion dollars, too. And so I [51:40] think what's happening is that they are, [51:42] again, there's they're better run [51:43] companies, no doubt, than the companies [51:45] in the '90s. That being said, like they [51:47] just can't help themselves. Like there's [51:49] it's it's just it's rational based on [51:51] game theory for them to be doing what [51:53] they're doing individually. Just [51:54] collectively in the [51:55] >> the multiples de-rate yet though. Mag 7s [51:57] are at an all-time high today, the [51:58] group. So what call what's the catalyst [52:00] for a re-rating? Well, I think the [52:01] challenge I think once the market [52:02] catches up to the perception that wait, [52:04] these are no longer the asset-light [52:05] businesses of 5 years ago, these are [52:07] utilities, right? Which is [52:08] >> And very tangible. Yeah, very capital [52:10] intensive with massive depreciation. [52:12] Right? That's the other thing, which is [52:13] you know, Chamath Palihapitiya on the [52:15] podcast, he was saying, "Look, like in [52:17] it's just mechanistically in any of [52:18] these um up cycles, what happens is you [52:21] have companies spending on CapEx, but [52:23] CapEx is um an capital asset on a [52:25] balance sheet that gets depreciated over [52:26] time. So for a 5-year depreciation, [52:29] you're only spending for each dollar [52:30] that's been spent on buying data [52:32] centers, you're only spending like 20 [52:33] cents each year hits your net income, [52:35] whereas Nvidia gets to record the entire [52:37] dollar as revenue. Um and so you end up [52:40] with this kind of just again, [52:41] accounting-based inflation of of net [52:43] income. So Michael and I had this [52:45] discussion 2 days ago. We were talking [52:47] about um [52:49] the the CapEx [52:51] just CapEx in general not really being [52:54] great for shareholders. [52:56] Um and this is of course like the [52:57] biggest CapEx cycle we've seen. [53:00] Um [53:02] and the example that I I don't know if [53:03] it's a good example or bad example, so I [53:04] want your opinion on this. [53:06] The example that I used is like AT&T and [53:08] Verizon. [53:09] The amount of money that they spent from [53:11] 1G to 5G just 25 years of endless [53:15] billions of dollars [53:17] these stocks are the same price they [53:18] were in 1997. Right. For I'm like I'm [53:21] not even not even exaggerating. Verizon [53:24] >> Josh, you're being paid to wait with the [53:25] dividend though. [53:26] Yeah, the dividend's not even the [53:27] dividend's not even great. Verizon, I [53:29] think it was still called Bell Atlantic [53:32] in 1998 was $45. It's $45. And here [53:36] here's [53:36] >> know there's a yield on it, but like [53:38] the the my I guess my question is [53:41] the these companies won. [53:44] The entire wireless business is three [53:45] companies. It's T-Mobile, [53:47] um it's uh Verizon, and it's AT&T. And [53:51] that's it. [53:52] >> Right. And what what did we what did we [53:54] win? [53:55] These stocks don't go up. [53:57] Um great. Uh [53:58] 5% dividend yields. All right, so it's [54:00] not zero. It's a bond. [54:03] So, it's not clear to me that this is [54:05] materially different. I'm pretty sure [54:07] Amazon, Alphabet, uh Oracle, I'm sure [54:10] they're going to win. But what do we win [54:12] as shareholders? So, that's my question [54:14] for you. What are your thoughts? Yeah, I [54:16] mean a couple of things. So, first of [54:17] all, if you look at the CapEx to sales [54:18] ratios, for many of the hyperscalers, [54:20] it's higher than AT&T at the height of [54:22] the dot-com boom. Right. [54:23] >> Right. So, they're certainly in the same [54:24] category. Um and yeah, it's a pyrrhic [54:27] victory. You can win the market, but [54:29] it's is it a market you want to win? [54:31] Because if your margins are like 60% [54:33] doing search, why do you want to get [54:34] into this other business? [54:35] >> tell you one I'll tell you one better. [54:36] You know who won? Apple. [54:38] Like uh Apple is the beneficiary of the [54:41] CapEx but the combined CapEx spending of [54:44] Sprint, which got sucked up into [54:45] T-Mobile, AT&T, and Verizon. [54:48] Apple doesn't own any of the networks. [54:51] They built the best product on top of [54:52] the network. Right. So, it's not that [54:55] CapEx bad, CapEx good. It's that who [54:57] what layer where do the profits accrue? [55:00] Now, we know in wireless, the profits [55:01] accrue to the iPhone Right. and the iOS [55:03] ecosystem. We don't know who's going to [55:05] win like really win this. [55:07] >> So, where are you looking for winners [55:08] then? [55:09] >> comes down to barriers to entry. It [55:10] comes down to the competition. Right? If [55:12] you're in a sector where it's easy for [55:13] people to compete, if your only moat is [55:16] oh, I have more money to throw at the [55:17] problem, well then like you know, I [55:18] could call up Masayoshi or whatever and [55:20] and solve for solve for that. So, that's [55:22] not really a a true moat. And that's why [55:24] historically capital-intensive [55:25] businesses have not actually been like [55:27] the best place to invest from an ROIC [55:28] standpoint. What our true moats are the [55:30] network effects you mentioned, these [55:32] intangible assets. You know, [55:34] no amount of money could have replicated [55:35] Google search um at the time. Now, it's, [55:38] you know, arguable that maybe AI will [55:39] help uh obsolete that technology, but at [55:41] least [55:41] >> of my searches are starting on Claude. [55:43] >> Yeah, and I think I think that is a [55:45] meaningful risk to to that part of the [55:47] business. But again, like there's the [55:48] question around the disruption of the [55:49] existing business. Microsoft is a good [55:51] example of a company that's, you know, [55:52] right in the crosshairs here. And then [55:54] the question of they're going into this [55:55] new business, will they win? And even [55:57] so, do they want to win? There are [55:58] people who think the profitability will [56:00] accrue to the layer that Palantir sits [56:02] on. [56:03] Where Palantir basically serves the top [56:05] of all of these LLMs and is the company [56:08] that gets paid by other companies to [56:11] literally tell them how to make best use [56:14] of all all this technology. [56:16] >> this? You have this quote here from Mo [56:17] and I don't know where this is from. He [56:18] said, "I think you would have to argue [56:19] that almost all the profit ends up going [56:21] to to the consumer ultimately and that's [56:23] because of competition." So, what would [56:24] what would that what would that look [56:25] like if if you're saying all the utility [56:27] goes to consumer? [56:28] Yeah, I mean it'll look like the [56:29] railroads where all the railroad [56:31] companies went bust and, you know, but [56:34] that being said, a ton of GDP was [56:35] created by the railroads. I have this [56:37] chart here. Um It is chart Let's see. [56:40] Hold on. Just on 216.1 [56:42] >> 36. This is from [56:45] um Azeem Azhar at Exponential [56:46] View. This is a good one. Yeah, this is [56:48] a good chart. So, [56:49] uh the blue line shows cost of railroad [56:51] construction each year. The green line [56:53] shows how much money how much earnings [56:55] the railroad companies earned. [56:56] >> Oh, wow. [56:57] >> red line shows the contribution of [56:58] railroads to GDP. How do they calculate [57:00] that red that red line? [57:01] >> I don't know. You got to You got to [57:03] click on the link. [57:03] >> Okay. [57:04] Yeah, [57:05] >> [laughter] [57:05] >> they made it up. They made it up. Um but [57:07] I mean, look look, if you take this this [57:08] at face value, what you see is, you [57:10] know, uh the costs above the earnings, [57:13] you can see the little spikes. Um these [57:15] are the panics when all the railroad [57:16] companies went went bust. And then even [57:19] after all that work, at the end of the [57:20] day, yeah, they were making some [57:22] profits, but they're utility-like [57:23] profits. [57:24] >> made the money. Who made the money? It [57:25] was, you know, the [57:27] >> manufacture furniture in North Carolina, [57:28] put it on a boxcar, ship it to another [57:30] city and sell it. [57:32] >> That's right. So, they'll be like AI is [57:33] going to help business formation or [57:34] something. It's going to make it easier [57:35] for people to do everything. [57:36] >> Now, well, so here is what the [57:38] hyperscalers would argue. They would [57:39] say, [57:41] "Well, we're invested in the LLM layer. [57:43] Like, we have partnerships and like [57:46] literal literal investments in some [57:48] cases." So And they would say, "We're [57:49] going to get a piece of that, too." I [57:51] actually happen to believe that the LLM [57:52] layer is also at risk of [57:54] commoditization. Take a look at number [57:55] 29. [57:58] Um so so what this chart shows Oh my [58:00] god, what's going on here? So this is [58:02] what the chart shows is like the just [58:03] the the Y axis is like how good are the [58:05] models, right? Go back to November 22 [58:07] when ChatGPT was released. The [58:09] underlying model was GPT-3.5. You can [58:11] see that at that point the OpenAI had a [58:13] huge lead over their competitors. In [58:15] fact, they were like two orders of [58:16] magnitude higher. [58:17] >> Now, it's a horse race and the the lead [58:19] the leadership is switching back and [58:20] forth. [58:20] >> Switching, yeah, it was Gemini, then it [58:21] was Claude, then it was GPT-5.5. [58:24] >> is a donkey, apparently. [58:25] >> L- Llama that that didn't work out too [58:27] well, yeah. [58:28] That Those efforts didn't work. [58:30] Then you have DeepSeek and you have the [58:32] Chinese models. [58:33] >> one of these players is going to break [58:34] away in terms of capability, but but one [58:38] or two of them already have broken away [58:40] in terms of [58:42] um [58:43] critical mass, like the amount of users. [58:45] Like, I think we have a Coke and Pepsi [58:47] >> brand, essentially. [58:47] >> we have a Coke and Pepsi and Dr. Pepper, [58:50] basically, at this point. I think Gemini [58:52] uh arguably is Coke. [58:54] Say Claude is Pepsi. OpenAI is Dr. [58:57] Pepper. [58:58] I like I think the other way around. I [59:00] think that Claude Claude is Coke, maybe. [59:02] >> Claude? [59:03] Claude is Coke, already? [59:05] >> I think Claude is or maybe OpenAI is [59:06] still Coke. [59:07] >> OpenAI is still OpenAI is still Coke, [59:08] yeah. I think cuz the average person [59:10] knows ChatGPT. The average person [59:11] doesn't know Claude. [59:12] >> It's more of a brand [59:13] >> Yeah, but the average person is doing [59:14] more AI stuff with Google and doesn't [59:17] even realize it. Every search is now an [59:19] AI workload. So I think it's I think [59:22] Gemini is the number one [59:24] a- like AI in terms of usage. [59:27] People could they don't they're not [59:28] doing it on purpose. They're getting [59:30] Gemini results instead of search [59:31] results. [59:32] >> they so so Google has a has a special [59:34] advantage is they have distribution, [59:35] right? They they They're putting they're [59:37] putting Gemini in my email. [59:39] >> Whether I like it or not, it's in there. [59:40] >> It's built into an installed base that [59:42] is significantly larger than what Open [59:44] AI and [59:45] >> think that's bigger than Open AI right [59:46] now. I do. [59:48] Well, it can be It can be measured. [59:50] >> Depends on how you measure it. [59:51] >> Yeah, and it's a question also of like, [59:52] you know, also the quality of the model. [59:54] I think right now it's it's clear that [59:55] Gemini is in third place. They will [59:57] likely release a new model next week [59:59] um and we'll and we'll see. Kai, I have [60:00] a two-parter for you. So, what first [60:02] brought you to intangibles? Like what [60:03] attracted you cuz your entire your [60:05] entire um operation is built around this [60:07] idea. And then part two, what does all [60:10] of the spending do to the intangibles [60:13] inside your portfolio? [60:14] >> Yeah, so answering the first question. [60:16] So, I I used to work for a company [60:17] called GMO. Um we are you know, kind of [60:19] like AQR, one of the, you know, pioneers [60:21] in quantitative investing. Um one of the [60:23] bread and butter factors within um you [60:26] know, uh GMO and other firms is the [60:28] value factor. The idea that, you know, [60:30] you buy if you systematically buy stocks [60:31] that are cheap relative to say book [60:33] value [60:34] um and, you know, underweight those that [60:35] are expensive, that that [60:37] historically has earned excess returns. [60:39] >> Did you chance to say that you helped [60:40] them update some of their models? Like [60:41] what do you say quality? You kind of [60:42] helped them update it a little or bring [60:43] up to speed to Yeah, that was one of the [60:45] things I worked on when I was an analyst [60:46] there. [60:46] >> Like modernize the way they're capturing [60:48] that Yeah, look and and and I it's been [60:50] been many years. I'm sure they they've [60:51] done more upgrades over time. Um and [60:54] again, not specific to them, but it it's [60:55] it's been a challenging time for value [60:58] investors um as as as defined via these [61:01] systematic factors. Like the I had [61:02] Mobius on the podcast, he made like a [61:04] cheeky point. He was like, "Look, the [61:05] Fama-French value factor is this like um [61:07] academic um [61:09] uh factor that basically is an index of [61:11] of cheap value stocks shorting or [61:13] underweight expensive stocks. It made [61:15] money every year for like 80 years and [61:17] then it stopped making money in 1994 [61:20] when they published their paper, right? [61:22] Um so so so that's the thing. So [61:24] >> Arguably like when America Online came [61:26] out, but Yeah, or or maybe when [61:28] when we all got computers Yeah. Yeah. [61:30] And so that was a big question, which is [61:32] like, you know, we were in a tough place [61:34] because you're sitting there and you're [61:35] saying, all right, well, like this [61:36] beautiful idea, you know, that Ben [61:38] Graham coined, um you know, 100 years [61:40] ago with you know, security analysis [61:42] that Dude, it's buy low, sell high. It [61:44] makes perfect sense. It should work. No [61:46] one is arguing that buy low, sell high [61:47] doesn't make sense. The question is [61:48] against what intrinsic value? Yeah. [61:50] Right? So that that was always my [61:51] contention. It's It's not that value [61:52] investing doesn't make sense. By [61:53] definition, it should make sense if you [61:55] know what true value is. The problem is [61:57] that the way we were measuring value was [61:59] just obsolete, right? Because so many of [62:01] the traditional metrics are [62:02] backward-looking. They don't take into [62:03] account [62:04] >> Accounting stuff. [62:05] >> R&D. They don't take into account [62:06] advertising and marketing, human [62:07] capital. Um they're based on just [62:09] tangible assets that, you know, at one [62:11] time 100 years ago made a lot of sense, [62:13] but as the economy has transformed from, [62:15] you know, industrial to [62:16] information-based asset light, companies [62:18] like Google and Nvidia have come to the [62:20] forefront. So you consider yourself a [62:21] value investor still? Absolutely, yes. [62:23] Um just a different [62:25] >> things differently. [62:25] >> Yeah, exactly. Because look, it's it's [62:28] the challenge is that if we don't have [62:29] an anchor value, how are we going to [62:30] invest, right? I mean, there's there's [62:32] different ways to invest, of course, but [62:34] you know, I I I do think that being able [62:36] to kind of marry the two schools, the [62:37] the idea of [62:39] um you know, being a value investor, [62:40] having some kind of price discipline, [62:42] you know, buying buying bargains and [62:44] knowing when to kind of sell and take [62:45] profits, you know, is a really important [62:48] um stabilizer for markets and a really [62:50] important thing to be doing. Um but we [62:51] need to update our metrics, right? Like [62:53] think about Warren Buffett and and [62:54] Berkshire. I talk about this story a [62:55] lot. That, you know, he started off as a [62:57] Ben Ben Graham's like actual disciple, [63:00] and he, you know, bought Berkshire [63:01] Hathaway and a struggling struggling [63:03] textile mill. He was only buying [63:05] struggling things. [63:05] >> And yeah, and and for and it worked [63:07] okay. And but but looking back, you [63:09] know, after meeting Charlie Munger, he [63:10] talks about this. He's like, look, I'm [63:11] never going to do that again, you know? [63:13] And then he bought Coca-Cola and then [63:14] Apple, right? And and the the implicit [63:17] lesson there is that he added intangible [63:19] moats to his framework. And Graham would [63:20] have done that, too. He He used to buy [63:22] stocks that were worth less than cash. [63:23] Like, guess what? That doesn't exist [63:24] anymore. That's right. All those He [63:26] would have updated his I think by the [63:27] end he was saying like he's an indexer [63:28] essentially in the '70s, but all those [63:30] guys, they [63:30] >> And then he retired at Yucaipa. He said [63:32] like, "Yeah, this [ __ ] doesn't work [63:32] anymore." Yeah. Yeah, I want to say that [63:34] like 84% of Buffett's investments were [63:37] were purchased with a price-to-book [63:38] above one. So, in other words, he was [63:39] not a price-to-book investor. [63:40] >> So, he changed his mind over time, not [63:43] just not just because he met Charlie [63:45] Munger, but also like the world also [63:47] changed. [63:47] >> Yeah, the the world changed and he had [63:49] to evolve his his process with it [63:50] because as you point out, Ben, there [63:52] just weren't enough things to buy [63:53] otherwise. [63:53] >> Yeah. He sort of also was an intangibles [63:55] investor because he recognized the power [63:58] of brand. [63:59] >> Geico. Yeah. Um Coca-Cola. [64:01] >> He recognized the power of things like [64:03] uh American Express and Coca-Cola and [64:06] what the brand meant to the consumer [64:08] Right. before [64:09] anyone else was really talking about [64:11] that. That's right. [64:12] >> So, there's like a lineage from Buffett [64:14] to what you're doing with your [64:16] intangibles portfolio. [64:17] >> what I'm trying to do is to kind of do a [64:18] systematic version of that. Yeah. Right? [64:20] Like, trying to say, I think you know, [64:24] anyone with common sense can sit there [64:26] and be like, "Yeah, I think brands [64:27] matter. I think human capital matters. I [64:28] think this company has a really [64:29] interesting technology." The challenge [64:31] is how do we quantify that? And so, for [64:33] the longest time, quants were in this [64:34] tough place because we had, you know, [64:37] Chris Combistat, the traditional [64:38] databases, which were all structured [64:40] data. Right? In your In your intro that [64:42] you read for me, you mentioned [64:43] unstructured data. Unstructured data is [64:45] everything else, right? The information [64:47] in in um in patents, in trademarks, in [64:51] um company filings, in news, and analyst [64:53] reports, in you know, on Twitter. All [64:55] this information contains obviously very [64:57] valuable information on companies and on [64:59] their intangible assets. [65:00] >> bring order to it. [65:01] >> If you can parse it. And I think for the [65:03] longest time the challenge it was that [65:04] quants, we had linear regression, we had [65:06] a few tools in our tool belt, but they [65:08] were all optimized for a world where [65:10] structured data was what we had [65:11] available. Now with large language [65:13] models and natural language processing, [65:15] we're we're finally able to kind of [65:17] unlock this, you know, huge trove of [65:19] information where I think the the task [65:21] which at one time seemed, you know, [65:23] unattainable of trying to codify a [65:26] Warren Buffett-style approach is now on [65:27] the table. Not saying that, you know, [65:29] I've done it or or, you know, anyone has [65:31] done it yet, but I think that, you know, [65:33] we're increasingly moving towards a [65:34] point where through AI and all these [65:36] tools, a lot of the, you know, kind of [65:39] intuition that's baked into a [65:40] fundamental investment strategy can be [65:42] codified in in quant. [65:43] >> So, how do you quantify brand? Because [65:45] we're in a market right now where [65:48] I think the worst stock in the world is [65:50] Nike. [65:51] And Lululemon is number two. [65:52] >> two. Lulu might be uh follow Lulu might [65:55] be following it down the drain. Um [65:58] is there a way to [65:59] is there a way to parse the things that [66:02] might go into a calculation where you [66:04] could say [66:05] this forget about the stock price having [66:07] lost value. [66:09] The brand was losing value a year [66:11] before. [66:12] And like that was the signal to not be a [66:13] Nike. Like can can we do things like [66:16] that? Is it effective? Is that part of [66:17] your approach to By the way, these [66:19] companies are both down 76% from their [66:22] highs. [66:22] >> Yeah, equally. Yeah. Crazy. So like [66:26] those are apparel specific, but just [66:28] generally like how do you say that a [66:30] brand is either gaining or losing value? [66:32] >> Yeah, I mean absolutely. If you were [66:33] Peter Lynch or or someone you'd say, [66:35] "Hey, you know, talk to your friends who [66:36] are Lulu customers. Are you still buying [66:38] their products?" Right? There's a I'm [66:40] friends with Peter Lynch. Oh, there we [66:41] go. Nice. [66:42] You should ask him then what he would [66:43] say. I don't think he wants to see me [66:45] again, but I I interviewed [laughter] [66:46] him I interviewed him for for an hour. [66:48] >> That's awesome. He's a legend. And he [66:49] was telling these great brand stories [66:50] about walking into the supermarket and [66:52] seeing [66:54] like one brand being sold at the [66:56] register and another being sold on a [66:57] shelf and, you know, like he he was [66:59] doing this in a very analog way. [67:01] >> Right. Okay. Yeah, and I think [67:02] increasingly you can start to [67:03] systematize some of these things. Right, [67:05] social media data is an obvious thing. [67:07] You can go on Instagram and [67:09] and like what what's the tone of the [67:10] mention? Is it positive or negative? [67:12] Right, like is it amongst the right [67:13] people? Like there's a kind of in crowd [67:15] of cool people, are they, you know, [67:17] again, you can buy influencers, so you [67:19] got to be careful for manipulation as [67:21] you do with with all signals, you know, [67:22] companies can juice earnings, you know, [67:24] whatever. But if you're careful about [67:26] it, yeah, if you track all this [67:27] information, in theory, you can [67:29] certainly capture, um, you know, where [67:31] where the trends are headed before, um, [67:33] you know, ideally, uh, the the stock [67:35] price. [67:36] Uh, how do prediction markets factor [67:38] into this? You must be really excited [67:40] about that data because that is [67:42] like I mean, especially things that are [67:45] uh, very far afield from what would [67:46] normally be in an 8K or or or, you know, [67:49] some sort of like official filing. [67:51] There's a lot of opinions being [67:53] expressed there and you can quantify it. [67:54] Yeah, I I think it's a really important [67:56] and interesting source of data. Um, I'm [67:58] not using a ton of it, to be honest, [67:59] yet. I mean, I think cuz most of the [68:00] stuff I'm doing is at the company level. [68:02] Right, so it's very useful if you're [68:03] trying to do macro forecasting, who's [68:04] going to win the election? Obviously, [68:06] there's a lot of sports betting and and [68:08] crypto stuff, which is, you know, less [68:09] relevant there, but, you know, it is is [68:12] Nike going to make a comeback? I mean, [68:14] that's I don't know if there's a [68:15] contract on that. If there were, it [68:16] probably not be a very liquid one. I [68:18] don't even know how come like how would [68:20] you even I I wasn't One thing I've used [68:22] is [68:23] One thing I've used is [68:25] Google Trends. Google Trends is an [68:26] interesting source of data. Okay. You [68:28] can like look at what people are [68:29] Googling, like terms, like if they're [68:30] Googling, [68:31] you know, the Lululemon Align like pants [68:33] or whatever, that's like a potentially [68:35] good thing and if that's going down, [68:36] that's a little bit concerning. You have [68:38] two ETFs. [68:39] What's the difference between the two [68:40] and which one should Ben Carlson buy? [68:43] Um, Pitch it. He's sitting right here. [68:45] Get him to buy it. [68:47] Um, so yeah, the first ETF, um, is [68:49] US-based, the one I launched in 2021. [68:51] Okay. Um, it buys a portfolio of stocks. [68:55] >> The Intangibles ETF. [68:57] >> Yes, that one. Somebody else had to say [68:59] it. [68:59] >> Thanks, Josh. And D Tangibles? [69:01] D Tangibles? What's that one? Developed [69:03] Markets. Yeah. So international [69:06] intangibles. [69:06] >> That's right. Yeah. And so that the [69:08] second one follows the same exact [69:09] strategy, both of which their goal is to [69:11] buy stocks that are cheap relative to a [69:14] expanded definition of intrinsic value [69:15] that includes these intangible modes [69:17] too. So brand, human capital, IP, [69:19] network effects. Um you know [69:23] you know about like how the IPOs are all [69:25] going to join the major indices faster [69:28] than ever. That's right. You making [69:29] plans for that new world where you got a [69:32] SpaceX and it's in the S&P 10 days [69:34] later. [69:35] Probably similar thing will happen with [69:38] Anthropic. Um like how do how does your [69:41] work get affected by that? You kind of [69:43] have to play along with it. Yeah, if [69:45] it's in the index, it's in the [69:45] investment universe. I don't have [69:47] It's not a It's an absolute return fund, [69:49] so I don't need to have I don't have a [69:51] benchmark per se. I don't need to own [69:53] anything, right? We're trying to make [69:54] money over the long run. So to the [69:56] extent SpaceX is added to the index, [69:58] it's now investable, which is nice. [70:01] I doubt it'll be in the portfolio day [70:03] one, um just given how [70:04] >> Is it quantitative or any discretion or [70:06] you rules based completely? Completely [70:07] rules based. The idea would be if [70:09] there's something that looks weird that [70:10] doesn't accord with fundamental [70:11] intuition, I'll ask the question of why [70:13] the model doesn't is missing that and [70:15] try to adjust the model um in a way that [70:17] solves the problem not just for this one [70:18] case, but also moving forward. [70:20] >> Are companies like SpaceX and Anduril [70:22] and some of the things that were some of [70:23] the more exciting companies coming [70:26] are they intangible assets or are they [70:27] tangible assets or they're it's a [70:29] obvious everything's a mixture. That [70:31] definitely a mixture. Um I mean some of [70:33] these companies are a little bit more [70:34] physical, right? They're not pure [70:35] software companies. I mean like SpaceX [70:37] they launch rockets, but obviously, you [70:39] know, most of their their value is in in [70:41] their IP and I do I do think that, you [70:43] know, a lot of their value is just in [70:44] Elon Musk in this case. No, that's the [70:46] ultimate intangible That's [70:48] like this the ultimate intangible asset [70:50] is Elon Musk's [70:51] involvement. [70:52] >> yeah. Right. Um hard pivot. [70:56] Ben's book. Can we can we say some words [70:59] about it? Are you allowed to speak about [71:00] it? It's not like an ETF, right? [71:01] >> Yeah, so there's no compliance rules [71:02] here. All right. [clears throat] [71:03] Uh [71:04] I will say I told Michael the other day [71:05] I'm not good at sales or self-promotion. [71:07] Like Josh, you're a salesman. [71:09] Is that good? Yes. Okay. [71:11] >> Josh could take Josh could take a [71:13] ketchup popsicle in 90° weather and sell [71:15] it to a woman wearing white gloves. You [71:17] could do that. [71:17] >> Thank you. [71:18] >> That's not me. Okay. So I'll I'll make [71:20] one hard sell for the book. Okay. [71:22] There's never been a better book for [71:25] charts and tables than this book. I [71:26] counted cuz Chartr had me help me. [71:28] >> Wait, say it again. That's that's a bold [71:30] claim. There's 50 [71:31] >> There's 52 charts [71:33] >> charts in every chapter in the book. [71:34] >> charts and tables in this book. It's got [71:36] more charts and tables and data than any [71:38] book [71:39] and Chartr had me help me, so I'm giving [71:41] him all the credit. Uh yes. [71:43] >> Way to not sell the audio version. [71:45] What do you do? [71:46] >> version comes [71:47] >> the chart? [71:47] >> No, it comes with a PDF of all the [71:49] charts. They send it to you. That's what [71:50] I do. Right? Whose idea was that? Is [71:52] that right? Yeah. [applause] Okay. So [71:54] anyway, that's my And I got a couple [71:55] charts. I want to There's one chart I [71:56] want to run by you guys. So chart 50, [71:58] John. Let's do it. Okay, so I looked at [72:01] the worst days, the worst months, the [72:02] worst years in stock market history. [72:04] Okay? So I did one that looked at the [72:06] worst months in stock market history and [72:09] most of them are in the '30s, '40s. [72:12] There's the 1987 on there. [72:13] >> chapter one. Yeah, chapter one. [72:15] >> I I read this chapter this morning on [72:17] the way into into the city, so I know [72:19] exactly what you [72:20] >> believe you asked Nick for a quote and [72:21] not me. [72:22] >> [laughter] [72:25] >> The funny thing is Nick was actually on [72:27] the cover and they bumped him for [72:28] Morgan. [72:29] Uh [72:29] >> [laughter] [72:29] >> so I I feel bad for him. Wow. Poor guy. [72:31] What's Is Morgan's quote even hot? Let's [72:34] see. "Has mastered the art of Ben [72:36] Carlson has mastered the art of exposing [72:38] the few big topics that matter most to [72:40] investors." [72:41] Let me hear Nick. What's Nick's quote? [72:43] Nick's not even on the back? [72:45] Oh, man. [72:45] >> He says [72:46] Where'd Nick go? It's on the inside [72:48] cover. [72:48] >> So anyway, so here's what I want to [72:49] Here's my point I want to make. Talking [72:50] about faster cycles. So there's been six [72:52] times where the stock market has been [72:53] down 20% or more in a single month. [72:56] I think you could make the case and [72:57] again, a lot of these are in the 30s. [72:59] You could make the case that going [73:00] forward, instead of having these massive [73:02] long drawn-out crises, like hey, we're [73:04] down 60% over 3 years. [73:07] We're going to have these more air [73:07] pockets because information moves [73:09] faster, where we have these huge down 1 [73:12] 2 3 4-month periods, where it's like, [73:14] "Oh my god, the stock market was down [73:17] 25% in a single month." That's where I [73:19] think we're headed in terms of the speed [73:20] of these things. I think there's going [73:21] to be more air pockets. Speed running [73:23] the correction. Cuz I wrote a I wrote a [73:25] chapter on the Great Depression and [73:26] people keep asking me, "Do you think it [73:27] could happen again?" And I think we've [73:28] completely cut that left tail off. So, [73:30] the question is we've cut that tail off [73:31] because of fiscal policy, monetary [73:33] policy. Okay, you can't The risk don't [73:36] ever go away. What does that mean? I [73:37] think it could mean [73:39] we just get faster more of these things. [73:41] >> So, buy and hold gets harder? [73:43] Mate, don't you think that these cycles [73:44] just speed up? There's no way this stuff [73:45] is slowing down. And AI is going to just [73:47] add more [73:48] >> That is literally what has hap- has [73:50] happened. My contention is that's where [73:52] we're going. It's just cycles are going [73:53] to be Of course that's your contention. [73:56] Right? What do you think What do you [73:57] think about that? [73:58] >> as well go whale hunting. I think that's [73:59] right. Yeah. [74:00] >> [laughter] [74:00] >> Like everything else is faster, why [74:02] wouldn't the market's ability to process [74:03] bad news be faster? [74:04] >> happening faster. There's a lot more [74:06] trading, 24-hour trading. Like we're all [74:08] we're heading towards a a world where [74:10] yeah, everything will be instant on your [74:11] phone. Right. There's no rule that says [74:13] how long a bear market has to be. I know [74:15] the I know the old heads like to say the [74:18] average bear market is 13 months. We [74:19] haven't had We haven't We haven't had a [74:21] a true economic recession [74:25] in a long time. [74:27] We really haven't. We haven't had a [74:28] credit cycle. [74:29] >> 18 years. We haven't had a credit cycle. [74:31] We'll have We'll have We'll have another [74:32] one. All right, I got one more chart. I [74:33] want to spike the football on some [74:34] people's heads. John, throw up chart 53 [74:36] in Japan. [74:38] This is the question I get more than any [74:39] other. Now show Japan. Meaning Meaning [74:41] the The implication is [74:43] >> Why are people so obsessed with Japan? [74:44] You Cuz Cuz Ben says it pays to be a [74:46] long-time investor, and people like, [74:48] "Oh, yeah? Japanese people sat in a down [74:51] stock market for three decades." [74:52] >> one I've never seen people show this. [74:54] From Yes, from 1990 to 2024, Japan did [74:57] nothing. It was like 1.5% per year. You [74:59] wouldn't know It was [75:00] Japan peaked in 1989 in like December [75:02] 1989. It bottomed in March of 2009. [75:05] Which is crazy. Horrible. [75:07] But, if you extend it, the returns were [75:09] so good in the 1970s and 1980s [75:12] that it was so compressed, it had to be [75:14] bad. So, from 1970 to 2024, you got [75:17] almost 9% in Japan. Long-term investing [75:19] did work in Japan. It was just all those [75:21] returns were compressed in the the first [75:24] Yes. [75:24] >> So, you did 22% a year from 1970 to [75:27] 1989. Small-cap stocks in Japan did 30% [75:30] per year for two decades. So, then you [75:32] do So, then you do 1% a year from 1990 [75:34] to 2024, which is a lifetime, let's be [75:36] honest. [75:37] >> average worked out. You still did okay [75:38] over the very long term in Japan. Over [75:40] the 60 Over the full 60-year period. [75:43] That's the cycle. [75:44] >> Right? [75:45] It's a good way of thinking about it. I [75:47] think most I like that. I like that. I [75:49] think most people would have preferred [75:50] if the the if the returns were back-end [75:52] loaded rather than front [75:54] >> [laughter] [75:54] >> rather than front-end loaded. All right, [75:56] dude, the book is called Risk and [75:57] Reward. I know Kai is uh very excited to [75:59] read it. Um you're going to you're going [76:01] to listen to it. [76:02] >> Uh I'm exhausted. Is Ben going to read [76:04] it to you? Yeah. I'm going to We're [76:06] going to call each other tonight. [76:06] >> I'm going to listen to it, 100%. I have [76:08] the audiobook, too. I'm going to listen [76:09] to it in bed. All right. So, I'm I'm I'm [76:11] reading it cuz I'm old school, but I [76:13] love it. I So, you know I'm like one of [76:15] the biggest fans of your writing in the [76:16] world, and I make it through the first [76:19] two chapters, and I'm just like, "Yeah, [76:21] man, this is what I need. This is the [76:22] medicine." Because it's all about things [76:25] will probably be okay. [76:27] It's It's a unique Most people writing [76:30] financial books, it's this the dollar is [76:33] going to not be the reserve currency [76:35] anymore, or gold is going to replace [76:37] You're just saying like, "No, no, no, [76:38] things will be okay." The 10% And here's [76:40] how you know. The 10% per year over the [76:42] last 100 years [76:43] >> inclusive of all the bad [ __ ] that's [76:44] happened. Right. [76:45] >> The Great Depression, that's part of it. [76:47] Yeah. [76:47] >> Right? 1987, that's part of it. 70s, [76:49] that's part of it. All the bad stuff [76:50] that's happened is inclusive in the [76:52] long-term returns that are still good. [76:53] That's the point. [76:54] >> been through a lot and things are still [76:55] okay. Yes. And I that's a really great [76:57] message. I love it. We should end there. [76:59] Guys, did you have fun on the show [77:01] today? [77:01] >> [applause] [77:02] >> Great time. Yeah? Kyle, you brought the [77:03] heat. Thank you. [77:05] You're so smart. Why are you so smart? [77:08] Yeah, another time. I want to tell [77:09] people where they can learn more about [77:11] your research cuz your research is [77:12] really spectacular. Your funds are [77:14] great, but like you're a thinker, you're [77:16] a philosopher, and you test your ideas [77:18] with data, and I I love what anytime [77:21] your stuff drops. Where do people go to [77:23] learn more about Sparkline and your [77:25] work? Well, thanks, Josh. Um yeah, you [77:27] can just go to my website, [77:28] sparklinecapital.com. [77:29] >> sparklinecapital.com. [77:31] Okay. And you're active on you [77:33] tweets? I tweet sometimes. I I I try to [77:35] respond. This guy's so smart, he doesn't [77:37] blog, he white papers. [77:39] Right. He's writing white papers while [77:41] we're doing blogs. Yeah, no doubt. All [77:43] right. Guys, thank you so much for [77:44] listening. Great job to the crew. I know [77:47] you guys worked your asses off this [77:49] week. John, Duncan, Rob, amazing. [77:52] Nicole, Daniel, [77:55] Travis, [77:56] >> [applause] [77:57] >> happy birthday to Graham Thomas. [78:00] Again, everybody. [78:01] >> [applause] [78:02] >> Thank you. [78:03] All right. Guys, thank you so much for [78:05] listening. We'll see you soon. [78:09] >> [music]