Altimeter's Gerstner Says Anthropic Cash Flow Anchor Spurs AI Rally

Altimeter的Gerstner称Anthropic现金流支撑推动AI涨势 Brad Gerstner认为，Anthropic创纪录的企业端收入增长和高毛利率证实了AI热潮的合理性，使大盘避免了10%至15%的回调。 投资者Brad Gerstner认为，AI繁荣从根本上得到了强劲企业需求以及Anthropic等基础模型领军企业爆发式收入的支撑。他警告不要实施拟议的数据中心暂停建设令，强调国家经济增长与安全取决于计算能力的扩张，并探讨了超大规模云服务商与私募市场如何应对Token限制。此外，Gerstner还详细剖析了“特朗普账户”（Trump Accounts）这一里程碑式立法倡议的推出，该两党倡议旨在为数百万美国儿童提供标普500指数的种子投资，以建立长期的资本所有权。 推理阶段计算（Inference-time reasoning）与自主智能体间通信的到来标志着AI市场的关键转折点，开辟了一条全新的扩展路径。 (1:04) Anthropic收入的快速增长和强劲的毛利率使市场情绪从对AI的怀疑转向看多，验证了企业对智能技术的需求。 (2:57) 企业对API使用和Token支出的优化并不会削减整体需求；相反，新AI用例的涌现和落地推广将轻而易举地超过效率提升带来的节省。 (5:18) 软件市场正在走向分化：直接处于Token流通链路中的数据基础设施（如Snowflake、Databricks）因此获益，而前端SaaS则面临护城河防御性风险。 (8:50) 美国地方暂停数据中心建设的举措，在与中国的AI竞赛中，对GDP增长、经济韧性和国家安全构成了生存维度的威胁。 (13:21)

Altimeter Capital's Brad Gerstner argues that current AI demand is physically token-constrained rather than a speculative bubble, driving capital into foundational compute and hardware infrastructure while leaving traditional application software increasingly vulnerable. To sustain this momentum, Gerstner contends that technology leaders must address local resistance to data center construction while expanding broad-based asset ownership through initiatives like the Invest in America Act. Physical Bottlenecks and Surging Token Demand (17:48) Gerstner rejects comparisons between the contemporary AI buildout and the 2000 dot-com crash, noting that unlike the unused dark fiber of the telecom era, there is no such thing as a dark GPU today because incremental compute is immediately utilized. In his assessment, the industry is constrained by tangible physical limits—specifically logic wafers, memory wafers, and powered shell data center capacity—rather than soft enterprise demand. He points to Anthropic's rapid revenue trajectory toward positive free cash flow and Dell's AI server business scaling from a $1 billion to a $16 billion run rate as concrete evidence that real cash flows underpin compute spending. Addressing claims that enterprise optimization will depress usage, Gerstner cites Altimeter's survey of 300 enterprises showing that firms actively optimizing API spend still project greater than 50% year-over-year spend growth, while those planning optimization expect approximately 90% growth. With OpenAI and Anthropic scaling from 3 gigawatts to a projected 10 gigawatts by year-end, Gerstner argues that expanding use cases in coding and knowledge work will outpace efficiency improvements. Software Bifurcation and Token Flow Investing (8:50) Gerstner observes a structural division across enterprise technology between companies positioned within the token flow and legacy software vendors. Database infrastructure providers like Snowflake, Databricks, and ClickHouse directly benefit as higher token consumption drives database queries. Conversely, front-facing software applications such as Salesforce risk being bypassed by foundational models. Gerstner places generic legacy software in the 'too hard' category, emphasizing that while software trades at 22–23x GAAP earnings, Nvidia trades near 13x forward earnings despite 70% growth. This dynamic has led to capital drying up for Series B and Series C software startups, prompting Altimeter to concentrate private holdings in frontier model creators (OpenAI, Anthropic) and hardware infrastructure (Cerebras, Groq). Gerstner also casts doubt on traditional service firms attempting in-house software development, such as Kirkland & Ellis allocating $500 million to custom systems, favoring instead private equity buyouts that rebuild professional service firms directly around frontier AI architectures. Data Center Expansion and Community Pushback (15:11) With hyperscalers committing over $100 billion annually to capital expenditures, Gerstner explains that peak-load dynamics are driving new enterprise offerings, citing Elon Musk's launch of Elon Web Services (EWS) with Anthropic as an anchor client and Meta's push into enterprise hosting. However, the physical footprint of these facilities is generating local friction, such as protests in Mishawaka, Indiana, regarding electrical grid strain and municipal water usage. Gerstner warns that data center moratoriums would trigger an immediate U.S. recession and hand strategic leadership to China, comparing local opposition to historical activism that curtailed domestic supersonic transport and nuclear power. To address these concerns, Gerstner is coordinating a coalition involving cloud providers, chipmakers like Nvidia and AMD, off-takers, and the White House to establish a direct 'community dividend' model, intended to provide host communities with tangible economic returns during a three-year transition toward widespread AI-driven abundance. Generational Wealth Through Trump Accounts (30:38) Beyond AI infrastructure, Gerstner outlines the rollout of 'Trump Accounts' under the Invest in America Act, which aims to provide equity ownership to the 35 million American children under age 10. The initiative grants a $1,000 baseline investment in the S&P 500 to children born after January 1, 2025, alongside $250 for children aged 2 to 10, supported by a $6.25 billion philanthropic commitment from Michael and Susan Dell. State and private contributions, including Gerstner's own matching grants in Indiana and North Carolina, are designed to stack on top of these baselines. Gerstner contrasts this direct-to-child mechanism with traditional philanthropic organizations, arguing that direct asset ownership eliminates administrative overhead. Assuming a $1,000 starting balance with $50 monthly contributions, Gerstner calculates an individual account could grow to approximately $50,000 by age 18. Over a 15-year horizon, he projects the initiative could transfer $3 trillion to $4 trillion in wealth to historically non-asset-holding households, with automated registration for the estimated 3.7 million annual newborns scheduled to begin in 2027. Gerstner's strategy relies on the thesis that physical compute constraints will continue to drive outsized returns in infrastructure rather than generic application software. However, the viability of this expansion hinges on whether proposed community dividends and federal wealth-building policies can effectively counteract local resistance to data centers and shifting state-level tax environments.

Altimeter Capital 创始人 Brad Gerstner 认为，当前市场由 AI 推理与算力基础设施的硬性需求驱动，资本正集中涌向处于“Token 流”（Token Flow）核心的底层硬件与模型层，而传统软件及后期初创企业则面临严峻分化与估值重构。 推理拐点与算力物理瓶颈 (1:04) Brad Gerstner 指出，本轮 AI 行情的真正转折点在于推理时间计算（inference-time reasoning）的突破，这开辟了智能扩展的新维度。随着自主 Agent 之间直接交互，英伟达 CEO Jensen Huang 预计推理需求将实现数亿倍的指数级爆发。在需求端，根据 Altimeter 对 300 家企业的独立调研，即使企业积极优化 API 使用效率，其预估的 Token 支出增速仍超过 50%，计划优化的企业支出增速甚至达到约 90%，表明效率优化并未抑制总体需求。在供给端，当前市场与 2000 年互联网泡沫时期的“暗光纤”（dark fiber）闲置完全不同，全球不存在闲置 GPU 或暗 Token，算力上线即被全额消化。整个生态受制于内存晶圆、逻辑晶圆以及带电数据中心机房（powered shell）等硬性物理瓶颈，OpenAI 和 Anthropic 的算力规模正从年初的合计 3 GW 向年底的 10 GW 及次年的 20 GW 推进。 企业分化与软件估值重塑 (8:50) 软件板块正在经历深刻的结构性分化。Gerstner 解释称，直接嵌入 Token 消费链条的数据库技术（如 Snowflake、Databricks、ClickHouse）因调用量激增而直接获益；相反，以 Salesforce 为代表的传统前端应用层 SaaS 面临被底层大模型直接绕过或替代的风险。目前传统软件公司的交易倍数已回调至约 22 至 23 倍 GAAP 市盈率，反映出市场对软件长期护城河的担忧与折价。此外，对于柯克兰·艾利斯律所（Kirkland & Ellis）计划耗资 5 亿美元自研内部自动化软件的举动，Gerstner 持强烈怀疑态度，认为非科技企业自研软件很难对抗专业大模型，更看好 Thrive Holdings 通过收购会计师事务所并嵌入顶尖 AI 工程能力进行改造的私募股权整合模式。 资本配置脱节与公私市场温差 (26:47) 资本市场呈现出公募与私募阶段严重割裂的景象。在二级市场，尽管英伟达过去三年股价上涨 15 倍，但其利润增速跑赢股价，前瞻市盈率降至约 13 倍，处于十年来的估值低位；美光、SK海力士等存储芯片厂商的市盈率也处于个位数。Gerstner 建议英伟达将自由现金流回报率提升至 70% 至 75%，以复制苹果吸引巴菲特等价值投资者的路径。与二级市场聚焦 AI 基础设施形成鲜明对比的是，传统中后期一级市场软件融资陷入冻结，过去从 A 轮几百万美元营收扩张到 B 轮 2000 万美元即可引发争抢的软件公司，如今在 B 轮和 C 轮极难寻得投资方。Altimeter 自身已将重心高度聚焦在 OpenAI、Anthropic、Cerebras、Groq 等基础设施与底层大模型标的上。 数据中心社会阻力与社区红利计划 (13:21) 针对部分地区叫停数据中心建设的提议，Gerstner 认为这将对美国经济扩张和国家安全构成灾难性威胁，可能导致类似当年超音速客机与核能发展因过度抵制而停滞的局面。针对印第安纳州米沙沃卡（Mishawaka）等基层社区对水电消耗与生活成本上涨的担忧，Gerstner 透露他正联合云厂商、英伟达、AMD 以及白宫牵头一项倡议，计划在未来三年的过渡期内向承接数据中心建设的社区直接发放实质性红利，以弥合社会政治分歧，争取公众对算力基建的支持。 全美投资法案与早期复利账户落地 (30:38) 围绕长期财富积累，Gerstner 详细介绍了基于《全美投资法案》（Invest in America Act）落地的“特朗普账户”（Trump Accounts）项目。该项目覆盖全美 3500 万名 10 岁以下儿童，2025 年 1 月 1 日后出生的婴儿将获得 1000 美元标普 500 指数种子基金，2 至 10 岁儿童则由戴尔夫妇基金会等慈善资金出资 250 美元建仓，地方与个人捐赠还可叠加匹配。配套移动应用上线后迅速升至全美应用排行榜第三位，资金将于 7 月 4 日全面就绪。Gerstner 预计该机制将在 15 年内为零资产家庭转移 3 万亿至 4 万亿美元财富，通过零行政磨损的直接股权持有，让下一代在 18 岁时积累约 5 万美元的基础资产。 Gerstner 的论述展现出对算力基础设施与美国资本扩张的强劲信心，但其描绘的未来高度依赖于大模型商业化变现的持续性，以及社区红利方案能否有效化解地方对数据中心高耗能的抵制。随着中后期传统 SaaS 融资持续承压，生成式 AI 的生产力转化效率仍有待更广泛的企业实践检验。

## Background 本视频于 2026 年 6 月 23 日由科技商业频道 TBPN 发布在 YouTube 平台上。节目记录了著名投资人、Altimeter 创始人布拉德·格斯特纳（Brad Gerstner）做客直播的深度对谈，围绕 AI 繁荣的早期演进、Anthropic 与英伟达等企业的商业化落地、中美算力竞争与数据中心建设，以及他牵头推动的“特朗普账户”（Trump Accounts）全民儿童持股倡议展开。 ## TL;DR 格斯特纳在访谈中深入剖析了 AI 产业格局，指出当前全球 GPU 与物理算力极度受限，与 2000 年互联网泡沫时代的“暗光纤”过剩存在本质区别，而传统软件板块正围绕是否身处“Token 流”出现剧烈分化。他痛批叫停数据中心建设的主张将摧毁美国经济并在对华竞争中拱手让出主导权，并倡导科技巨头为地方社区建立切实的分红机制。在二级市场投资上，他坚定看好估值被严重压缩的英伟达、存储芯片巨头以及前沿模型开发者。此外，他重点拆解了已正式立法落地的“投资美国法案”与儿童账户计划，阐述如何通过国家红利、戴尔夫妇等超级富豪的零损耗慈善匹配，让全美儿童借助标普 500 的终身复利实现从 0 到 1 的财富跨越。 ## Atmosphere & style 整场对话节奏紧凑明快、信息密度极高，在宏观资本博弈的硬核推演与人文关怀之间自如切换。格斯特纳与主持团队展现出深厚默契与插科打诨的幽默感，开篇关于播客出售的戏谑与收尾处互换角色的调侃极大地活跃了现场气氛。而在深入探讨算力国家安全、亲赴印第安纳农村学校见证贫困家庭因拥有资产而落泪的段落中，格斯特纳的情绪饱满且极具感染力，让整期节目呈现出一种兼具实干洞察与强烈技术乐观主义的昂扬基调。 ## Bulletin - [03:04](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=184s) Anthropic 凭借惊人的商业化增速与高毛利率支撑了今年美股的 AI 叙事，其第二季度自由现金流有望转正，避免了整体股市可能出现的两成回调。 - [08:58](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=538s) 软件板块正在依据是否身处“Token 流”发生严重分化：Snowflake 等底层数据赋能者查询量激增，而 Salesforce 等前端应用则直接面临底层大模型的能力替代与估值压缩。 - [18:05](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1085s) 与 2000 年互联网泡沫时期的“暗光纤”闲置截然不同，当今全球没有任何闲置的 GPU 或 Token，AI 扩张正受限于物理晶圆与通电数据中心机房的硬性供给约束。 - [30:36](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1836s) 《投资美国法案》落地并推出官方 App，通过联邦注资与 Michael Dell 等慈善家配捐为全美儿童设立长期复利账户，预计 15 年内实现 3 至 4 万亿美元的代际财富转移。 ## Key takeaways - [01:02](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=62s) “推理期思考”扩展了智能演进的新向量，黄仁勋预判在多智能体（Agent-to-Agent）协作驱动下，推理算力需求将呈 10 亿倍级爆发。 - [03:04](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=184s) Anthropic 展现出极高增长与毛利率并有望实现自由现金流转正，挽救了资本市场对 AI 商业化兑现的疑虑，成为今年 AI 核心叙事的定海神针。 - [05:17](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=317s) 针对 300 家企业的独立调研显示，即便在积极优化 Token 开支的企业中，未来一年预算增速仍超 50%，有力驳斥了“Token 极大化”需求崩塌论。 - [08:58](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=538s) SaaS 软件板块正经历结构性分化：紧扣 Token 流的数据库基础设施（Snowflake、Databricks）迎来业务激增，而传统前端 CRUD 软件直面大模型侵蚀。 - [10:28](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=628s) SaaS 估值调整仅回落至美股大盘的平均基准（GAAP 市盈率 22-23 倍），多数估值倍数仍显著高于英伟达，抄底传统软件极易沦为接飞刀。 - [13:17](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=797s) 呼吁暂停数据中心建设极其危险且短视，不仅会立即重创依赖 AI 驱动的美国 GDP 增长，更会将 AI 产业与国家安全主导权拱手让给中国。 - [15:04](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=904s) 为缓解地方社区对水资源和电费上涨的抵触，科技巨头正联合白宫设计“社区红利”机制，搭建渡过未来三年关键过渡期的社会政治桥梁。 - [18:05](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1085s) 与 2000 年宽带基数小且光缆大量闲置的互联网泡沫不同，如今全球拥有数十亿互联人口，行业正处于严重的物理算力供给受限状态。 - [21:36](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1296s) 千亿级资本支出倒逼超大规模云巨头面临“AWS 难题”，Meta 进军企业级 AI 代理与马斯克对外出租算力云（EWS）均源于峰值闲置算力的商业化外溢。 - [24:35](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1475s) 传统律所自研底层软件胜算极低，未来更有前景的路径是由 Thrive Holdings 等资本直接收购传统服务机构并用一线技术团队进行 AI 工作流重构。 - [27:06](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1626s) Altimeter 聚焦投资 Token 流核心枢纽，规避 50 亿至 150 亿美元估值的成长期软件公司，并将大量资源集中于 OpenAI、Anthropic 及二级市场算力基建。 - [28:55](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1735s) 二级市场算力资产呈现反常低估，英伟达在 70% 增速下估值倍数仅约 13 倍，随着自由现金流大比例用于回购分红，将迎来类似苹果当年的投资者结构蜕变。 - [30:36](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1836s) 《投资美国法案》通过全民儿童账户让每个孩子生来拥有标普 500 资产，构建直达个人、无中间管理抽损的“捐赠誓言 2.0”基建。 - [36:02](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2162s) 从 2027 年起全美新生儿将凭社会安全号自动建账，以 1,000 美元本金和每月定投复利，在 18 岁时积累数万美元，从产权机制上超越传统福利体系。 - [39:10](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2350s) 迈克尔·戴尔夫妇注资 62.5 亿美元覆盖 2,500 万儿童，树立了直达受众、高度弹性的现代慈善新标杆，彻底规避传统基金会 30% 的行政损耗。 - [41:40](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2500s) 加州是全美甚至全球经济的关键阵地，必须坚决击败具有违宪没收性质的财富税提案，通过公投确立财产保护法案以稳固创新根基。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=0s) 开场调侃：SpaceX 预期与被收购传闻 节目一开始，主持人与 Brad Gerstner 热情打招呼，并提到刚才 Kyle 刚在节目中给出了极具震撼力的 SpaceX 看多逻辑——“谁掌控了太空，谁就掌控了世界，因此必须重仓 SpaceX”。主持人顺势将问题抛给 Brad，希望他能进一步深入剖析：在 SpaceX 走向 IPO 的进程中，他在重点观察什么？有哪些悬而未决的问题？又有哪些常人尚未意识到的市场误解？ 面对直奔主题的追问，Brad 笑着打趣道，大家连一点“前戏”都没有就直接切入硬核的 SpaceX IPO 话题。他随即反客为主，调侃起主持团队近期把播客/业务卖掉的传闻：“如果你们真把这东西卖了，我之后还见得到你们吗？我感觉我们现在是在为你打工呢。”他甚至戏谑地追问团队在出售时是不是拿到了 OpenAI 的股份、到底赚了多少钱。现场在一阵欢声笑语中完成了破冰，主持团队也笑称大家现在都在“同一条船上”，随后将话题引向更宏观的技术与市场演变。 ### [01:02](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=62s) 突破疑虑之墙：从推理期思考到 Agent 爆发 主持人将记忆拉回到此前两人在 Jeffrey Katzenberg 活动上的线下碰面。当时市场正处于一段微妙的“迷你回调期”：聊天机器人（Chatbots）经历了最初的爆发式增长后热度有所降温，而 AI Agent（智能体）才刚刚开始展露雏形。整个市场在第四季度经历了犹疑与震荡，但 Brad 当时就精准预判了 Agent 的崛起，宛如手握“水晶球”。 Brad 坦言，在过去两年半的时间里，市场其实经历了好几次这样的微型调整，投资界始终横亘着一座“疑虑之墙”（Wall of Worry）。他提到在自己与投资传奇 Bill Gurley 共同主持的播客中，两人曾反复激烈辩论： - 真实的 AI 商业化营收到底能不能兑现？ - 算力投入能否带来健康的毛利率（Gross Margins）与真实的投资回报率（ROI）？ - 业界是否在盲目过度建设（Overbuilding）？历史上每一次供应链紧缺是否最终都会演变成严重的产能过剩（Glut）？ ```mermaid flowchart TD accTitle: AI推理扩张与Agent飞轮演进 accDescr: 展现从推理期思考到多智能体交互引发的计算需求与产业突破过程 subgraph Scaling["智能扩展新维度"] A["预训练规模瓶颈"] --> B["引入推理期思考 (Inference-time Reasoning)"] B --> C["突破智能临界点 (如 Opus 4.6)"] end subgraph AgentFlywheel["Agent 交互与算力爆发"] C --> D["具备实质业务价值的 AI Agents"] D --> E["Agent 与 Agent 协同交互"] E --> F["推理算力需求呈指数级爆发 (10亿倍增长)"] end subgraph MarketImpact["商业化验证"] F --> G["企业级实际用例落地"] G --> H["打破市场疑虑之墙"] end ``` 在 Brad 看来，打破这一疑虑周期的关键转折点，正是“推理期思考”（Inference-time reasoning）的出现，它为模型智能的扩展开辟了一条全新的向量。Nvidia 首席执行官 Jensen Huang（黄仁勋）曾在做客 Brad 的播客时斩钉截铁地表示：“Brad，推理算力的需求增长绝不是 100 倍或 1000 倍，而是将增长 10 亿倍（1 billion x），因为未来的核心场景是 Agent 与 Agent 之间的直接对话与协作。”进入 12 月初，随着 Opus 4.6 等前沿模型的问世，模型智能明显跨越了一个核心阈值，展现出截然不同的实用价值，预示着全新增长周期的确立。 ### [03:04](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=184s) Anthropic 拯救叙事与算力硬件狂飙 年初开局时，整个资本市场依然对 AI 商业化收入的兑现程度抱有极大的怀疑。Brad 明确指出，如果今年没有 Anthropic 拿出惊人的收入业绩，“整个股票市场可能会下跌 10% 到 15%”。这并非夸大其词，而是因为无论是 OpenAI 还是 Google，其近期的商业化数字虽然表现良好，但并未达到“彻底震撼市场”的超预期水平。真正作为核心引擎力挽狂澜、撑起整个 AI 叙事与估值体系的正是 Anthropic——它堪称“资本主义历史上增长最快的公司”。 更为关键的是，Anthropic 不仅收入增速破纪录，还维持了极高的毛利率，甚至传出在第二季度有望实现自由现金流（Free Cash Flow）转正的积极信号。这一信号彻底提振了整个 AI 板块，扭转了两个月前大盘年内平盘甚至微跌的阴霾。Brad 透露，Altimeter 旗下公募基金刚刚经历了其 18 年历史中最强劲的两个月回报期。 这种强劲的现实需求同样在硬件供应链中得到了淋漓尽致的印证： - **戴尔（Dell Technologies）**：Brad 的好友 Michael Dell 带领公司交出了惊人的答卷，其 AI 服务器业务收入同比暴增 750%，规模从原本的 10 亿美元骤增至 160 亿美元。 - **美光（Micron）**：股价从早前的每股两三百美元直接飙升至千元水平。 - **戴尔股价**：从一年前的 80 至 90 美元一路攀升到 400 美元附近。 Brad 强调，这些惊人的涨幅证明技术落地是实打实的。但他也提醒投资者保持理性：要想让这种繁荣持续，必须依赖消费者持续愿意付费的使用量、大中小型企业真金白银的支出增长，以及主权 AI（Sovereign AI）领域的采购。在漫长的上升通道中，阶段性收入不及预期、半导体股票出现 10% 至 20% 的常规技术性盘整，都是再正常不过的健康调整。 ### [05:17](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=317s) “Token 极大化”之争与企业优化真相 针对市场上一旦出现利好便有人寻找做空借口的现象，主持人抛出了当前行业的核心争论：在看到 Anthropic 惊艳收入的同时，空头开始质疑是否存在虚幻的“Token 极大化”（Token maxing）泡沫，即企业是否在缺乏明确 ROI 的情况下盲目狂烧 Token？ Brad 剖析了硅谷当前极度撕裂的两种声音： 1. **持续悲观派**：此前断言 AI 根本不会产生商业收入；如今收入摆在眼前，又改口称这些收入全都是泡沫与骗局，认为全是在无脑堆砌 Token、毫无 ROI 可言。 2. **极端乐观派（AI Pilled）**：坚信当前一切消费都已达到帕累托最优（Pareto optimal），坚称每家企业每一分钱的 Token 开支都极其合理且无可挑剔。 Brad 认为真相在两者之间。正如 Altimeter 自身在购买调用 Token 时绝不会故意浪费钱一样，上百万个独立的商业决策主体都在基于自利进行理性决策。为了厘清事实，Altimeter 专门针对 300 家跨行业企业展开了独立调研，调查企业在优化 AI 支出的背景下对未来 12 个月预算的规划： | 企业当前对 AI 支出的态度 | 未来 12 个月 API / Token 支出预期增速 | |:--- |:--- | | **已在积极优化现有支出（Actively Optimizing）** | 预计同比增长 **> 50%** | | **正计划开展支出优化（Planning to Optimize）** | 预计同比增长 **~ 90%** | 调研数据直接击碎了“优化支出就会导致需求崩塌”的空头逻辑。Brad 分析指出： - 尽管企业在实践中必然会追求算力使用效率并剔除低效实验，但当前技术渗透率依然极低。 - 代码生成（Coding）作为目前最成熟的场景，企业也仅仅处于刚上车的起跑阶段。 - 更广泛的知识工作者（Knowledge Work）场景几乎还未真正铺开。 - 全球范围内尚有绝大多数企业根本还没开始使用 AI。 因此，即使伴随着持续的使用优化，由于底层渗透率提升与场景扩张的增长曲线极其陡峭，Anthropic 和 OpenAI 仍将穿越所谓“优化周期”，持续保持强劲扩张。就在讨论深入之际，主持人抛出了下一个直指核心的问题：这是否是一个零和博弈的市场？在 Token 上花费的每一美元，是否都在直接削减传统 SaaS 软件公司的预算？ ### [08:58](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=538s) 软件股分化：身处 Token 流（Token Flow）与面临大模型竞争的区别 在经历了所谓的“SaaS 末日”（SaaS-pocalypse）重创之后，软件板块近期出现了一些反弹信号，主持人询问 Brad Gerstner 如何看待这一波行情。Gerstner 提到，他前一天登上 CNBC 时就以 Snowflake 为例：该公司单日大涨约 35%，不过拉长时间来看，其年内涨幅仅约 10%，远逊于美光（Micron）或 ARM 等年内暴涨 200% 的硬件半导体企业。在他看来，软件行业正在发生深刻的分化（Bifurcation），市场过去习惯把所有 SaaS 软件公司一锅端、等量齐观，但现在真正的分水岭在于**公司是否身处“Token 流”（Token Flow）之中**。 Gerstner 解释道，像 Databricks、Snowflake 和 ClickHouse 这类数据基础设施公司（Altimeter 均有投资），其业务逻辑与 Token 消耗紧密绑定。用户消耗的 AI Token 越多，产生的数据库查询量就越大。以 Altimeter 自身的运行情况为例，其数据库查询量的增速甚至超过了 Token 使用量的增长。这类公司已经证实自己处于 Token 价值流动链条上，因而重新获得了 AI 估值溢价。相反，像 Salesforce 这样的前端应用软件公司面临着截然不同的困境。尽管 Gerstner 极为推崇 Marc Benioff 的商业手腕，但他指出，Salesforce 提供的很多前端解决方案与底层基础大模型本身存在竞争关系。Snowflake 是大模型的“赋能者”（Enabler），而 Salesforce 却更容易被模型直接替代或挤压，因而面临更艰巨的挑战。 - **板块年内表现对比**：Snowflake 年内仅涨约 10%（即便财报后单日大涨 35%），而半导体标的如 Micron、ARM 年内涨幅高达约 200%。 - **核心分化机制**：身处 Token 流中的数据库基础设施（Snowflake、Databricks、ClickHouse）随着 Token 消费增加而查询激增；前端应用软件（如 Salesforce）则直面大模型的能力侵蚀。 ```mermaid flowchart LR accTitle: AI时代软件公司的价值分化逻辑 accDescr: 展示Token消耗如何对底层数据赋能者带来业务激增，同时对前端应用软件形成竞争替代压力。 subgraph UserLayer ["AI 交互与推理层"] direction TB TokenUse["用户端 Token 消耗增加"] GenAIModels["基础大模型 Capabilities 提升"] end subgraph Enablers ["底层赋能者 (处于 Token 流中)"] direction TB DBQueries["数据库查询量激增 (Snowflake / Databricks / ClickHouse)"] MultipleExpansion["获得 AI 估值溢价倍数"] end subgraph Competitors ["前端应用层 (直接竞争)"] direction TB AppSoftware["传统前台应用 (如 Salesforce)"] MultipleCompression["面临大模型替代 / 估值折价"] end TokenUse --> DBQueries --> MultipleExpansion GenAIModels --> AppSoftware --> MultipleCompression ``` --- ### [10:28](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=628s) SaaS 估值重置真相与巴菲特的“太难”投资篮子 Gerstner 回顾了关于“软件已死”的讨论：早在一年半前，他与投资人 Saqja 录播客时，对方提出“软件不过是构建在 CRUD 数据库之上的单薄用户界面（thin UI on top of a crud database）”，随即引发了包括 Benioff 在内整个行业的恐慌与抗议；随后 Gerstner 又与 Bill Gurley 就“软件是否已死”展开辩论。市场在 12 月份普遍陷入恐慌并大幅下调估值，但 Gerstner 拿出一张精心准备的对比图表，指出关键在于**这些估值到底重置到了什么水平**。 数据显示，这次估值修正仅仅是将软件板块从远高于市场平均水平的高估值状态，拉回到了美股大盘的平均基准——即包含真实股权激励（SBC）在内的 GAAP 市盈率约 22 至 23 倍。Gerstner 强调，大部分软件股的估值倍数依然高于英伟达（Nvidia）。英伟达作为 AI 领域最不可或缺的核心底座，在业绩增速高达 70% 的情况下，市盈率仅约 13 倍，而许多增速放缓的软件公司估值倍数反而是英伟达的两倍。因此，所谓的“软件估值冤枉受损”根本站不住脚：市场只是将其过去“坚不可摧”的护城河预期下调，提高了折现率。他警告称，未来如果软件业务跟不上 AI 智能提升的步伐，估值完全有可能跌破大盘均值，下方仍有下跌空间。援引沃伦·巴菲特著名的“简单、困难、太难”（too hard basket）分类法，Gerstner 直言当下的传统软件板块对他而言正处于**“太难看懂”（Too Hard Basket）**的篮子里，抄底极易演变成“接飞刀”。 - **当前估值基准**：SaaS 股票包含真实 SBC 的 GAAP 市盈率回落至 22-23 倍，与标普大盘均值持平，并未被严重低估。 - **与英伟达对比**：英伟达在 70% 的爆发式增长下市盈率仅约 13 倍，而增长乏力的软件股估值倍数高达其两倍左右。 - **投资评级倾向**：对于自认为能战胜市场的抄底资金，由于企业若无法融入 Token 流就将面临进一步估值压缩，传统软件整体属于“太难搞懂”类别。 --- ### [13:17](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=797s) 数据中心暂停令的狂妄，以及对全美经济与中美竞争的灾难性后果 针对近期市场上关于可能出台“数据中心建设暂停令”（Data Center Moratorium）的传闻，主持人提出：如果新增算力受限，虽然硬件供应链受挫，但现存拥有 Token 销售能力的公司是否会因议价权提升而受益？Gerstner 斩钉截铁地反驳：“这对所有人都是坏事，而最严重的打击将落在美利坚合众国身上。” 他告诫硅谷切勿盲目自大，历史教训历历在目：当年正是一小撮激进环保活动人士彻底扼杀了美国的超音速民航客机技术，同样也是一小群活动人士叫停了美国的所有核电清洁能源建设——导致如今中国正在兴建 100 座核反应堆，而美国仅有 1 座在建。Gerstner 情绪激动地指出，那些呼吁暂停数据中心建设的“疯子们”完全没有意识到，当前美国大部分 GDP 增长正是由数据中心建设、AI 推进以及由此带来的经济生产力提升所拉动的。一旦推行数据中心暂停令，其直接后果将是： 1. **美国经济即刻陷入衰退与高失业率**； 2. **在 AI 竞争中一夜之间将全球主导权拱手让给中国**。这场竞争远不仅关乎算法，更是经济安全、本土就业和国家安全的终极决战。 --- ### [15:04](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=904s) 印第安纳乡镇的现实焦虑：凝聚科技巨头与白宫的“社区红利”方案 虽然极力反对叫停数据中心，但 Gerstner 坦承地方社区的忧虑完全合情合理。他分享了刚在印第安纳州乡村为母亲庆祝 90 岁大寿的经历：在诸如印第安纳州米沙沃卡（Mishawaka）这样的小城镇，当地居民最操心的是自己和孩子的饭碗，而外来激进分子跑到镇上危言耸听，宣称数据中心会耗尽当地水资源、推高居民电费账单。在这种情况下，普通民众产生抵触与焦虑完全情有可原。毕竟在当地人眼里，自家后院建起的“AI 工厂”除了带来短暂的施工岗位和少许维护工种外，并无直接好处，而 AI 成果全世界都能享用。 为了解决这一深层矛盾，Gerstner 透露他正牵头一项尚未正式公布的重大举措：联合全产业链核心参与者——包括各大云巨头、英伟达、AMD 以及核心用电大户（Off-takers），并协同白宫，设计一套向数据中心落户社区输送“非常切实现实且深远的分红”（profound dividend）的机制。主持人盛赞他是“美国资本主义应得的英雄”，继推动全民持股倡议“特朗普账户”（Trump Accounts）之后，这正是他的下一场重头戏。 Gerstner 强调，未来 3 年必须建立起坚实的“社会-政治桥梁”（socio-political bridge）。三年之后，AI 为普通消费者带来的巨大丰足感将不言自明——每个人口袋里都将拥有一个近乎免费的个人超级助理，管日程、叫外卖、买衣服、给母亲送生日礼物无所不能；企业效率也将全面升维。正如约翰·梅纳德·凯恩斯在工业革命初期一样，Gerstner 坚定地站在“技术进步乐观派”阵营，但也坚决拒绝在眼前的过渡期阵痛面前当驼鸟。唯有让基层民众在未来 3 年内尝到切实利益，才能确保全美算力基础设施建设平稳跨过阻碍，不被逆风颠覆。 ### [18:05](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1085s) 从 2000 年暗光纤到算力紧缺：为何 AI 不是当年的互联网泡沫 当被问及当前技术突破日新月异、企业端与消费端边界不断模糊，内部是否在采用新框架来评估产品推向市场的速度时，布拉德·格斯特纳（Brad Gerstner）回顾了自己入行的历程。他在 1999 年至 2000 年涉足科技投资，当时全美乃至全球仅有约 3500 万宽带上网人口。虽然大家早早看清了亚马逊的未来潜能，但当年投资者犯的最大错误就是“跑得太快、过于超前”，忽略了底层的宽带受众只有区区几千万人。反观今天，全球已有 30 亿至 40 亿互联人口，技术的扩散速度与渗透规模已不可同日而语。 更根本的区别在于当下的物理限制与供需格局。布拉德指出，如今 AI 的扩张存在一道天然屏障：全球的内存晶圆、逻辑晶圆以及通电的数据中心机房（powered shell）产能都是有限的，这意味着能够生产出来的 token 总量极其有限。当年铺设光纤网络时，行业创造了“暗光纤”（dark fiber）这一名词，因为大量光缆埋入地下时大家心知肚明根本无人使用；而“当今世界上没有任何一块闲置的 GPU，也没有任何一个闲置的 token”。从各大科技巨头的最新财报来看，谷歌明确表示自己受到了“token 约束”，并坦言只要有更多 token 就能创造更多营收；亚马逊、微软、OpenAI 以及 Anthropic 面临的境遇完全一致。人类对智能的需求在爆发，智能的产出受制于物理层面的算力产能。 这种强劲的需求拉动让布拉德坚信未来几个月将迎来惊人的技术飞跃。他提到，今年年初 OpenAI 与 Anthropic 两家拥有的算力合计仅为 3 吉瓦（GW），而到今年年底这一数字将逼近 10 吉瓦，明年底更有望达到 20 吉瓦。伴随算法改良和沿着缩放定律（scaling laws）的大步迈进，类似 Cursor 团队（由迈克尔·特鲁埃尔 Michael Truell 等人领衔）这样在一线打造前沿模型的创业者，或是 OpenAI 与 Anthropic 的核心成员，眼神中都带着一种“奥本海默般的神情”，坚定地传达出“我们已经走到这一步了”的震撼。布拉德强调，自己并不担忧短期内偶尔某个月份收入增速放缓的“泡沫论”，他最关切的是美国必须在算力与模型竞争中全速前进、赢得全球 AI 竞赛，用国家层面的巨大财富积累来拉高整个社会的经济底线。 - **宽带时代对比**：2000 年互联网宽带用户仅约 3500 万，目前全球互联网人口已达 30 亿至 40 亿，技术扩散的基数差异巨大。 - **暗光纤 vs. 活跃 GPU**：2000 年铺设的大量光纤空置未用；如今算力市场“没有暗 GPU，也没有闲置 token”，全行业处于 token 供给受限状态。 - **算力飞跃数据**：OpenAI 与 Anthropic 年初合计算力仅为 3 GW，预计今年年底接近 10 GW，明年年底逼近 20 GW。 --- ### [21:36](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1296s) 千亿资本支出与“AWS 难题”：Meta 进军企业级与马斯克的算力生意 针对近期报道称 Meta 正在积极招聘企业端团队、准备切入企业市场的消息，主持人感到意外，因为 Meta 历来在消费端拥有无与伦比的优势——数十亿级活跃用户与硬件生态。布拉德对此解释称，一旦一家公司的年度资本支出（capex）迈向千亿美元门槛，它就不可避免地会遭遇经典的“AWS 难题”。当年杰夫·贝佐斯（Jeff Bezos）创办亚马逊云科技（AWS），起因正是亚马逊必须按照黑色星期五和圣诞购物季的最高峰值来配置服务器集群，而一年中其余时间这些昂贵硬件有一大半处于闲置状态，因此将其对外出租成了最优解，最终反哺了主营业务。 ```mermaid flowchart TD accTitle: The AWS Problem in Massive Capex Era accDescr: Flowchart illustrating how massive capex leads to surplus compute off-peak and drives consumer giants into the enterprise compute rental business. subgraph Investment ["大规模基础设施投入"] A["年度资本支出达 1000 亿美元"] --> B["按峰值需求建设庞大算力集群"] end subgraph Dilemma ["AWS 难题产生"] B --> C["非峰值时段硬件闲置成本极高"] end subgraph Monetization ["算力对外商业化变现"] C --> D["对外提供算力租赁与企业级云服务"] D --> E["埃隆·马斯克推出 EWS 并签约 Anthropic"] D --> F["Meta 拓展企业级 AI 代理与算力服务"] end ``` 这种由庞大基建催生的供给外溢同样发生在埃隆·马斯克身上。马斯克顺理成章地推出了所谓的“EWS”（Elon Web Services，即马斯克算力云服务），并且 Anthropic 已成为其核心客户。布拉德评价道，放眼当今世界，“没有人比埃隆更擅长把电力（电子）转化为 token”，未来马斯克的数据中心不仅会遍布地球，最终还会开进太空，这种算力布局也从根本上提升了外界对 SpaceX 潜在估值与 IPO 预期的兴奋度。 回到 Meta，在首席财务官李苏珊（Susan Li）的精细运营与马克·扎克伯格誓死不退的 AI 雄心之下，Meta 不可能放弃前沿大模型的军备竞赛，因此他们必须将建成的庞大算力资产全面变现。虽然从一家 120% 纯正的消费级互联网公司转型去做 AWS 模式或企业级 AI 代理难度极高，但如今通过代码生成等工具实现的“产品驱动增长”（PLG）模式，本质上带有非常强烈的消费级基因。加之 Meta 的广告平台本身就与全球数十万乃至数百万家企业保持着直接业务往来，Meta 在企业级市场的发力很可能会超出传统投资者的预期。 --- ### [24:35](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1475s) 律所砸 5 亿自研软件的迷局：Thrive Holdings 赋能模式与 AI 时代的并购整合 访谈转向了一则行业动态：全球顶尖律所柯克兰-埃利斯律师事务所（Kirkland & Ellis）据报道正计划投入高达 5 亿美元自研内部管理与业务软件。针对传统非软件机构投入数亿美元造软件屡屡以失败告终的历史教训，主持人询问布拉德，这一轮 AI 代码生成革命是否会促使更多专业服务巨头选择自研整套技术栈，以摆脱对 Harvey 或 Leya 等新兴法律 AI 软件的依赖。 布拉德直言，律所做出这种表态某种程度上是迫于竞争压力的无奈之举，“否则难道要向市场宣布直接放弃吗？”但他明确表示，如果自己是柯克兰的合伙人，绝不会赞同这一提案，因为由一群日常主业是撰写法律诉状的专业人士去自建团队，试图开发出媲美甚至超越 OpenAI 和 Anthropic 能力的杀手级软件，概率微乎其微。 相反，布拉德更看好资本与技术跨界整合的路径，例如好友乔希·库什纳（Josh Kushner）旗下的 Thrive Holdings 所采取的模式： 1. **收购传统服务业企业**：直接出资收购会计师事务所等传统专业服务机构。 2. **顶尖技术团队下场改造**：由精通工程的顶级技术人员与 OpenAI 建立深厚合作，直接深入业务流程底层重构工作流。 3. **实现实质性生产力飞跃**：连 OpenAI 联合创始人格雷格·布罗克曼（Greg Brockman）都在社交平台上转发称赞其为会计行业带来的巨大效率红利。 布拉德预判，未来私募股权机构（PE）和类似 Thrive 的机构将会掀起大批对传统服务企业的私有化收购潮，通过“AI 超级加持”来实现降本增效，而不是指望传统律所凭借一己之力在软件研发上战胜硅谷的科技巨头。随后，话题自然延伸至一级市场投资环境，主持人就当前风险投资机构在 A、B、C 轮成长期的出手节奏与观望心态展开了进一步追问。 ### [27:06](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1626s) 拥抱“Token 流”与 AI 算力短缺的投资哲学 主持人谈到在早期投资阶段，投资人很难看清未来究竟哪些公司会被时代滚滚向前的车轮无情碾压，并询问 Brad Gerstner：在当今这种押注未来 10 年结果的极早期阶段，置身于“Token 流”（Token Flow）之中，是否正是他感到最安心、最愿意出手下注的地方。Brad 坦诚地回应道，在当今这种时刻，所有人都必须保持谦逊，因为看清未来 10 年的走势几乎是不可能的，“甚至看清未来 10 个月都极其困难”。不过，回顾 Altimeter 创立 18 年以来的历程，目前公司的表现比以往任何时期都要出色，早期投资团队的战绩尤其惊人。 谈到具体的投资标的，Brad 指出核心就在于紧扣“Token 流”，围绕算力短缺展开布局。比如 Altimeter 持有 Cerebras 长达 9 年并迎来了其 IPO，同时他们还投资了 Groq，并在半导体、算力数据中心等领域持续深耕。此外，他们还重点布局了诸如军事现代化等 AI 相邻且深度受益于 AI 的赛道。Brad 强调，必须寻找那些“要么身处 Token 流之中，要么能从 Token 流中获益”的公司。而在成长型投资方面，他们基本避开了估值在 50 亿至 150 亿美元之间的拐点期（inflection stage）软件公司，而是将资金集中下注在 OpenAI 和 Anthropic 这两家正在消耗数十亿美元的领军者身上，将它们视为技术红利的核心最终受益者。 ```mermaid flowchart TD accTitle: Altimeter AI 投资全景架构 accDescr: 展示 Altimeter 在 Token 流、基础设施、模型层和二级市场的全方位布局 subgraph TokenFlow["核心算力与 Token 流基础设施"] A["半导体与芯片<br/>(Cerebras, Groq)"] B["算力与数据中心基建"] C["AI 衍生赛道<br/>(军事现代化等)"] end subgraph FrontierModels["前沿大模型独角兽"] D["OpenAI"] E["Anthropic"] end subgraph PublicMarket["公开市场价值低估核心资产"] F["Nvidia<br/>(估值倍数处于十年低点)"] G["存储芯片巨头<br/>(SK Hynix, Micron - 个位数市盈率)"] end TokenFlow --> FrontierModels TokenFlow -.-> PublicMarket ``` - **早期与成长期分化**：成长阶段的中间层（50 亿至 150 亿美元估值）投资被大幅压缩，资金集中于像 OpenAI 和 Anthropic 这样真正决定格局的顶级巨头。 - **软件市场融资现状**：过去软件行业里 Series A 做到几百万美元营收、Series B 做到 2000 万美元营收就能引发机构排队抢份额的盛况已彻底消失，如今同类软件公司在市场上“连一个愿意接盘的人都找不到”。 --- ### [28:55](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1735s) 公开市场的超常回报与 Nvidia 的估值洼地 在公开市场配置上，Altimeter 已经连续 3 年将几乎 100% 的投资组合集中在 AI 与算力赛道。Brad 抛出一个让许多人惊讶的事实：过去三年里，甚至超越风险投资（VC）级别超高回报的超级机会，实际上就发生在公开二级市场。他举例指出，SK Hynix 和 Micron 的交易倍数依然停留在个位数，而哪怕过去三年股价暴涨了 15 倍的 Nvidia，目前的估值倍数也仅有 13 倍左右。 面对“暴涨 15 倍后估值何以如此之低”的疑问，Brad 解释其根本原因在于业绩与盈利爆发的速度远超股价扩张的速度，导致其估值倍数不仅没有泡沫化，反倒被大幅压缩，当前估值甚至处于 Nvidia 近 10 年来的最低水平。同时，Nvidia 正在将其自由现金流的 50% 用于股息分红和股票回购。Brad 直言他甚至想公开建议 Jensen Huang（黄仁勋）将回购和分红比例提高到 70% 或 75%。他做出预测：正如当年 Warren Buffett 在 Apple 承诺将 50% 至 75% 的自由现金流返还股东时大举建仓并创造了投资神话一样，一旦 Nvidia 迈过这个股东回报的资本配置门槛，它的市场地位与投资者结构将迎来全新蜕变。 --- ### [30:36](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1836s) “投资美国法案”落地：全民儿童账户与 Giving Pledge 2.0 针对主持人体恤当下父母焦虑、询问下一代资产配置与最新政策进展的问题，Brad 详细分享了他四年来全力推动的成果。去年 7 月 4 日，作为重大法案一部分的《投资美国法案》（Invest in America Act）正式通过立法，并将于今年 7 月 4 日正式注资启动。更令人振奋的是，配套的官方 App 已经正式上线，全美拥有 10 岁以下孩子的家庭都应该立即下载注册。全美 10 岁以下儿童共有约 3500 万人，每人至少能获得 250 美元的初始资金；2025 年 1 月 1 日之后出生的婴儿（约 2 岁以下）将直接获得 1000 美元存入标普 500 指数基金。 ```mermaid flowchart LR accTitle: 全民儿童投资账户资金与注资机制 accDescr: 展示法案如何将国家、慈善家及各州资金注入儿童账户并投资标普500 subgraph Allocation["资金分配阶梯"] A["新生儿 (2025年1月1日后出生)<br/>$1,000"] B["2-10 岁儿童<br/>$250 基础注资"] end subgraph Matching["社会与地方匹配加码"] C["Michael & Susan Dell<br/>匹配 $250"] D["印第安纳州 (Brad Gerstner)<br/>额外 $250"] E["康涅狄格州 (Ray Dalio)<br/>额外 $250"] F["俄克拉荷马州政府<br/>额外 $250"] end subgraph Investment["终生复利池"] G["标普 500 指数 (S&P 500)<br/>复利锁定 18 年"] end Allocation --> G Matching --> G ``` Brad 将这套体系称为“捐赠誓言 2.0”（Giving Pledge 2.0）。未来美国将发生数以万亿美元计的代际财富转移，而这是资本家资助下一代最高效的途径：每一分钱都 100% 直达孩子名下，并在标普 500 中持续复利至少 18 年。数据证明，拥有此类资产账户的孩子高中和大学毕业率更高，未来创业和购房的概率也显著增加，其带来的整体社会投资回报率（ROI）难以估量。 - **上线成绩与热度**：App 发布仅仅一天便冲上全美 App Store 榜单第三名，排在前面的只有 ChatGPT 和 Claude，甚至超越了 Google。 - **跨界联手执行**：项目由 Robinhood 的 Vlad Tenev、BNY（纽约梅隆银行）、National Design Studio 的 Joe Gebbia（Airbnb 联合创始人），以及财政部团队（包括财政部长与 Luke Petitt）协同打造，成为“公民提出构想、国会立法、硅谷精锐突击队落地开发”的绝佳范例。 --- ### [34:03](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2043s) 重塑 70% 未被代表者的命运：从达勒姆试点到人人拥有资产 Brad 透露，今年 7 月 4 日资金将正式开闸注资，全美每个参与的父母都能在 App 中看到自己的孩子拥有了一小份 Nvidia、Microsoft、Walmart 以及全美前 500 强企业的股权。他希望在 7 月 6 日能在白宫椭圆形办公室联合敲响纽交所（NYSE）和纳斯达克（Nasdaq）的开市钟，以宣告这些账户正式开启交易。除了购买指数基金，Brad 还在力劝身边的顶尖科技领袖们（涵盖 Meta、SpaceX、OpenAI 等知名企业创办人），号召他们每家向全美每个孩子赠送一股自己公司的股票。 这项政策的核心目的在于彻底重塑那 70% 感到被体制遗忘、被资本抛弃的底层民众的命运。传统的 529 大学储蓄计划本质上只是前 10% 有储蓄能力的富裕阶层的特权，而该计划让全美每个孩子生来就成为国家资产的“拥有者”并搭上长期复利的列车。Brad 动情地分享了上周五他在北卡罗来纳州达勒姆（Durham）一所学校的亲身经历：他以个人名义认领了一所拥有 700 名学生的小学，该校 75% 的学生为黑人和拉丁裔，主要服务当地农村贫困家庭。通过一张包含 700 名孩子的 Google 表格，他向校长开具了 15 万美元支票，为全校每个孩子注入了 250 美元初始资金。现场一位母亲当场激动落泪，直言“我做梦都没想过自己的孩子有一天也能拥有资产”；老师们也热切期待着借此向孩子们讲授什么是资本、什么是真正的资产所有权。Brad 呼吁全美的成功人士行动起来，认领身边的学校，为下一代铺就走向财富复利的道路。 ### [36:02](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2162s) 从0到1的复利奇迹：特朗普账户与全民私有制构想 布拉德·格斯特纳（Brad Gerstner）回顾了自己在印第安纳州农村长大的经历。当时家庭资产完全是零，他向总统直言，处在“零”的境地是一种令人绝望的体验，因为身处其中的人根本不知道如何迈向“一”。“世界上最艰难的跨越就是从0到1；从1到2要容易些，从2到3更是如此。”格斯特纳指出，这项计划的核心就是让所有美国孩子完成从0到1的跃迁，开启财富复利的旅程。若以1000美元为起点，每月持续存入50美元，到孩子18岁时账户金额就能达到5万美元。适逢美国建国迈向第二个250周年，这一举措将作为面向全美儿童发放的国家红利落地。从2027年起，全美每年新出生的约370万名婴儿将实现自动建账——只要取得社会安全号码（SSN），就能自动获得一个专属的特朗普账户（Trump Account）。 格斯特纳不仅推动政策，也在自己的机构身体力行：他的公司约有35名员工、共计80个孩子，每年年底都会直接向每个孩子的账户存入500美元，甚至通过二维码就能直接将资金扫码转入。他呼吁各行各业的雇主都行动起来，不论是中小企业、大型集团、房产中介还是餐饮店，都可以为员工子女注资。该倡议搭建起了一个普遍私有制（Universal Private Ownership）的开源基础设施平台，让家庭真正拥有资产产权并体会到储蓄的尊严，相当于“为每位美国公民打造伴随一生的终身401(k)计划”。主持人对此感叹不已，坦言最初听到格斯特纳构想时，以为这属于“难如登天、几乎不可能做成”的事务，如今看到方案真正变成现实，简直是一剂最强劲的乐观主义解药（White Pill）。 - **起步储蓄模型**：1,000 美元本金起步，每月定投 50 美元，在复利驱动下，孩子年满 18 岁时可积累约 50,000 美元。 - **全自动建账机制**：计划自 2027 年正式生效，当年出生的 370 万名婴儿在申领社会安全号（SSN）的同时自动生成账户。 - **企业率先示范**：格斯特纳旗下公司约 35 名员工覆盖的 80 名子女，每人每年年底将收到 500 美元账户注资。 ```mermaid flowchart TD accTitle: Trump Account Compound Architecture accDescr: Flow diagram showing how automatic enrollment, multi-channel capital inputs, and long-term compounding work together for every American child. subgraph Enrollment ["建账与准入机制"] A["婴儿出生取得社会安全号 SSN"] --> B["自动开立特朗普账户 / Invest America 账户"] end subgraph Funding ["多方资金注资渠道"] C1["联邦国家红利"] --> D["终身复利账户"] C2["企业与雇主定投"] --> D C3["个人与家族储蓄"] --> D C4["慈善家定向捐助"] --> D end subgraph Compounding ["复利积累与未来应用"] B --> D D --> E["18岁成年积累：50,000 - 200,000 美元"] E --> F1["首套房首付款"] E --> F2["高等教育与创业启动资金"] E --> F3["持续滚入终身资产组合"] end ``` ### [38:05](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2285s) 3至4万亿美元的代际转移：超越社保的私人财富杠杆 格斯特纳强调，虽然目前该项目仍处于起步的第一天，但长远影响力不可估量。根据向总统汇报时的估算，未来15年内，该系统预计将推动3万亿至4万亿美元的财富，从富裕阶层转移至原本资产为零的人群手中。总统本人甚至认为这可能会成为其任内最重大的历史遗产。格斯特纳评价道，随着时间的推移，这项计划对社会的影响力甚至将超越传统的社会保障（Social Security）体系。两者的本质区别在于产权属性——它绝非政府救济项目，而是由公民个人真正拥有的私人账户，享有独立资产所有权，并在其漫长生命周期中持续产生复利。 主持人进一步补充了该机制对整体宏观经济的连带刺激效应：当整整一代年轻人步入成年、组建家庭时，账户里已经沉淀了5万、10万甚至20万美元，这足以成为购置首套住宅的首付款。这一坚实的购买力底座反过来能承销和激励更多房屋的营建，激活上下游市场力量。格斯特纳完全认同这一看法，坚信20年内由此引发的连带经济效应将极其深远。 - **财富转移规模**：15 年内预计实现 3 万亿至 4 万亿美元财富向零资产群体的定向赋能。 - **产权属性差异**：完全属于个人与家庭的私有产权，杜绝传统政府福利资金池的分配损耗。 - **宏观连带效应**：年轻群体手握大额购房首付款，从需求端拉动房地产建设，形成积极的经济正循环。 ### [39:10](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2350s) 迈克尔·戴尔的创纪录捐赠：彻底重塑传统慈善模式 格斯特纳提到，这一构想能从图纸变为现实，离不开关键盟友迈克尔·戴尔（Michael Dell）及其夫人苏珊·戴尔（Susan Dell）的鼎力支持。戴尔夫妇在最后关头助推政策完成了跨越终点线的“临门一脚”，并随即宣布了人类慈善史上规模最大的一笔直接捐赠：出资62.5亿美元，为全美2500万名儿童每人注资250美元。格斯特纳认为，戴尔夫妇的行动仅仅是个开始，为硅谷新一代超级富豪树立了前所未有的标杆。当下硅谷创造财富的速度与规模在人类历史上绝无仅有，但传统的慈善机构根本无力消化和承载100亿、200亿甚至500亿美元级别的巨额资本。 许多富豪希望在有生之年或离世后10年内散尽家财，且要求资金去向精准、可自由伸缩——既能包揽整个州，也能精准资助一个郡县甚至一所具体学校。与以往常常收取高达30%管理费、高管领取千万年薪且在捐赠者死后运作拖沓的传统基金会不同，这套新基础设施能确保100%的捐款分文不差地直接沉淀进孩子名下的账户中。主持人也对此深表赞同，直言旧时代的慈善概念过于空泛，富豪们宣布捐出一半财富，但实际资金滞留在繁冗的慈善壳子里迟迟未能落地，外界感受不到任何实质改变；而这种直达底层的账户体系从根本上颠覆了传统慈善游戏规则。 - **创纪录慈善注资**：迈克尔与苏珊·戴尔承诺出资 62.5 亿美元，按每人 250 美元的标准直接覆盖 2500 万名儿童。 - **零损耗管理机制**：告别传统慈善组织普遍存在的 30% 运营管理抽成与高管千万薪酬损耗，实现 100% 资金直达个人。 - **高弹性资助颗粒度**：富豪与捐赠机构可按全州、县域甚至单一学校等任意地理与组织维度定向进行认领式捐助。 ### [40:56](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2456s) 冲浪趣闻、硅谷家常与马克·安德森的转发 节目直播互动环节，观众在聊天公屏中询问格斯特纳近期是否有冲浪旅行计划。格斯特纳打趣道，这肯定是某位翻看了他社交媒体头像的网友提的问题。他澄清说，那张拍摄于冲浪农场（Surf Ranch）的标志性照片里，在浪管（Barrel）中穿梭的并不是他自己，而是他当时年仅11岁的儿子，当时冲浪专家雷蒙多（Ramondo）正在一旁现场指导孩子如何卡进浪管。格斯特纳笑言自己刚刚度过了55岁生日，眼下的工作强度比以往任何时候都要大，根本不可能抽出大块时间去休假冲浪，最多只能抽空去冲浪农场切磋一下，并热情邀请主持团队未来一同前往。 这种高强度的工作状态与他对事业的投入紧密交织在一起。尽管日常日程被密集的大型项目占满，他在家庭与工作中依然保持着充沛的精力与幽默感。 - **头像乌龙澄清**：社交媒体头像是在 Surf Ranch 拍摄的钻浪管照片，主人公是他 11 岁时的儿子而非他本人。 - **55岁的精力管理**：刚过 55 岁生日，处于人生中工作节奏最紧张忙碌的阶段。 ### [41:40](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2500s) 加州保卫战：击败财富税与重塑加利福尼亚的常识 话题随后转向加州的现状与未来政治经济走向。格斯特纳透露，他们正在为全美各城市的孩子寻找认领出资人，目前旧金山与奥克兰已全额落实，洛杉矶全域的儿童账户也即将敲定认捐者。谈及加州，他态度鲜明地表示绝不放弃这片土地，誓要击败被他称为“违宪没收税”（Unconstitutional Taking Tax）的所谓财富税或超级富豪税。他痛斥这种政策是在对成功进行污名化攻击、人为制造社会撕裂；而“特朗普账户”与“投资美国账户”（Invest America Accounts）的目标恰恰相反，是要抬高整个社会的底线，让所有人都能参与财富共赢，彻底扭转把埃隆·马斯克（Elon Musk）等领军企业家逼离加州的恶性局面。 格斯特纳还自豪地提到了自己正在读高中的儿子林肯·格斯特纳（Lincoln Gerstner）。林肯与斯坦福大学知名教授约书亚（Joshua）合作撰写了一篇探讨加州税收政策经济影响的学术论文，刚在网上公开发布。儿子回家时带着几分挑衅和得意地问父亲：“马克·安德森（Marc Andreessen）转推过你的文章吗？”在父亲确认未曾有过之后，儿子笑着炫耀安德森刚刚转推了他发表的论文。格斯特纳在节目中大方称赞了安德森的慧眼与对年轻一代的鼓励。 面对外界对加州前途的一片悲观论调，格斯特纳给出了鲜明的反向预判： 1. 他打趣预言斯宾塞·普拉特（Spencer Pratt）甚至可能成为洛杉矶新市长； 2. 财富税提案必将被彻底击溃； 3. 加州将通过公投正式确立《退休与个人资产保护法案》（Retirement and Personal Asset Protection Act），在法律层面彻底禁止任何侵害公民退休金与私人财产的行为。 格斯特纳指出，全美其他地区往往将加州视为无可救药的“深蓝州”，但加州骨子里其实是相当务实的“紫州”。常识终将在接下来的选举中重新占据上风。作为全球第四大经济体，加州的发展直接关乎整个美国的走向——“加州向何处去，美国就向何处去”。绝不能将加州拱手让人，必须在这片土地上捍卫符合立国精神的优秀理念并赢得胜利。格斯特纳还在节目中特别向谢尔盖（Sergey）以及“建设更美好的加利福尼亚”（Building a Better California）团队致敬，感谢他们为扭转加州发展方向所作出的卓越贡献。 - **城市儿童账户认领进展**：旧金山与奥克兰的资助人配对已完成，洛杉矶全境儿童即将由知名出资方全部认领完毕。 - **法案与公投攻防**：力推加州公投通过《退休与个人资产保护法案》，从法理根基上封死财富税与非法侵夺个人养老金的路径。 - **学术与家庭插曲**：高中生儿子林肯与斯坦福学者合作发布加州税改论文，引来马克·安德森的主动转发与点赞。 - **核心政治信条**：“加州向何处去，美国就向何处去”——加州是全球第四大经济体，绝不可轻易放弃其思想与制度阵地。 ### [44:23](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2663s) 尾声：互换角色的BG2玩笑与告别 在播客的最后几分钟，主持团队对格斯特纳的深度分享表示诚挚感谢，并对由他发起的一系列新项目寄予厚望。大家继续打趣冲浪的邀约，格斯特纳则笑着调侃称，自己打算把今天这场访谈变成他自己的播客节目《BG2》，反客为主来采访两位主持人，探探他们节目的底细与近况。伴随着轻松欢快的笑声与周末祝福，格斯特纳与主持团队互道珍重，结束了本期充实而高密度的对话。 ## Notable quotes > [01:02](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=62s) Brad，推理算力的需求增长绝不是 100 倍或 1000 倍，而是将增长 10 亿倍（1 billion x），因为未来的核心场景是 Agent 与 Agent 之间的直接对话与协作。 — Jensen Huang > [03:04](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=184s) 整个股票市场可能会下跌 10% 到 15%。 — Brad Gerstner > [10:28](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=628s) 软件不过是构建在 CRUD 数据库之上的单薄用户界面（thin UI on top of a crud database） — Saqja > [13:17](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=797s) 这对所有人都是坏事，而最严重的打击将落在美利坚合众国身上。 — Brad Gerstner > [18:05](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1085s) 当今世界上没有任何一块闲置的 GPU，也没有任何一个闲置的 token。 — Brad Gerstner > [21:36](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1296s) 没有人比埃隆更擅长把电力（电子）转化为 token — Brad Gerstner > [27:06](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=1626s) 甚至看清未来 10 个月都极其困难 — Brad Gerstner > [34:03](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2043s) 我做梦都没想过自己的孩子有一天也能拥有资产 — 达勒姆学校家长 > [36:02](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2162s) 世界上最艰难的跨越就是从0到1；从1到2要容易些，从2到3更是如此。 — Brad Gerstner > [41:40](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2500s) 马克·安德森（Marc Andreessen）转推过你的文章吗？ — Lincoln Gerstner > [41:40](https://www.youtube.com/watch?v=HRYysSAv-Ro&t=2500s) 加州向何处去，美国就向何处去 — Brad Gerstner ## Mentioned Brad Gerstner, Kyle, SpaceX, OpenAI, Jeffrey Katzenberg, Bill Gurley, BG2, Jensen Huang, Nvidia, Anthropic, Claude, Opus 4.6, Google, Altimeter, Dell Technologies, Michael Dell, Susan Dell, Micron, Databricks, Snowflake, ClickHouse, Salesforce, Marc Benioff, ARM, Saqja, Warren Buffett, Apple, John Maynard Keynes, Jeff Bezos, Amazon Web Services (AWS), Elon Musk, Susan Li, Meta, Mark Zuckerberg, Kirkland & Ellis, Harvey, Leya, Josh Kushner, Thrive Holdings, Greg Brockman, Cerebras, Groq, SK Hynix, Invest in America Act, Trump Accounts, Robinhood, Vlad Tenev, BNY, National Design Studio, Joe Gebbia, Airbnb, Luke Petitt, Walmart, Ray Dalio, Surf Ranch, Lincoln Gerstner, Marc Andreessen, Spencer Pratt, Building a Better California, TBPN ## Video description This is our full conversation with Brad Gerstner, recorded live on TBPN. We discuss why he believes the AI boom is still in its early innings, how companies like Anthropic, Nvidia, and Snowflake are benefiting from the AI infrastructure buildout, why America must continue investing in data centers and compute capacity to stay ahead of China, and his vision for Trump Accounts, a new initiative designed to give every child a stake in the American economy through long-term stock ownership. Sign up for TBPN’s daily newsletter at TBPN.com Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive [Source](https://www.youtube.com/watch?v=HRYysSAv-Ro)

[1:02] Jensen Huang: Brad，推理算力的需求增长绝不是 100 倍或 1000 倍，而是将增长 10 亿倍（1 billion x），因为未来的核心场景是 Agent 与 Agent 之间的直接对话与协作。 [3:04] Brad Gerstner: 整个股票市场可能会下跌 10% 到 15%。 [10:28] Saqja: 软件不过是构建在 CRUD 数据库之上的单薄用户界面（thin UI on top of a crud database） [13:17] Brad Gerstner: 这对所有人都是坏事，而最严重的打击将落在美利坚合众国身上。 [18:05] Brad Gerstner: 当今世界上没有任何一块闲置的 GPU，也没有任何一个闲置的 token。 [21:36] Brad Gerstner: 没有人比埃隆更擅长把电力（电子）转化为 token [27:06] Brad Gerstner: 甚至看清未来 10 个月都极其困难 [34:03] 达勒姆学校家长: 我做梦都没想过自己的孩子有一天也能拥有资产 [36:02] Brad Gerstner: 世界上最艰难的跨越就是从0到1；从1到2要容易些，从2到3更是如此。 [41:40] Lincoln Gerstner: 马克·安德森（Marc Andreessen）转推过你的文章吗？ [41:40] Brad Gerstner: 加州向何处去，美国就向何处去

[17:48] Brad Gerstner: When we were putting down the fiber in 2000, we called it dark fiber for a reason. There was nobody using it, and we knew there was nobody using it when we put it in the ground. There's not a dark GPU in the world today. [26:47] Brad Gerstner: This is the cheapest multiple Nvidia has traded at in a decade right now... all the venture returns have been had in the public markets, by the way. [38:05] Brad Gerstner: When you're at zero, it's a despondent place to be. You don't know how to get to one... We're going to get all of these kids from zero to one on this compounding journey.

[0:00] Welcome back. How you doing? [0:02] >> Great to be here, guys. Great to be [0:03] here. [0:03] >> Fantastic. [0:04] >> Yeah, it's back to back with Kyle. [0:06] >> Yeah. Yeah, he he laid out one of the [0:09] greatest SpaceX bull cases. We're going [0:11] to throw it to you to try and one up [0:12] him. He said, "Whoever controls space [0:14] controls the world, and for that reason [0:16] you got to own SpaceX." I liked it. But, [0:19] take us to the level deeper. What are [0:20] you thinking about in the SpaceX IPO in [0:22] the lead up? What are you watching for? [0:24] What unanswered questions are there? [0:26] What do you think is misunderstood maybe [0:28] that more people should be aware of? [0:29] >> I mean I mean come on, guys. We didn't [0:31] even have any foreplay and you're [0:32] already driving me down. You know, is [0:35] SpaceX IPO? I mean first, I haven't seen [0:37] you guys if you sell this thing if you [0:40] sell it. [0:41] >> Yeah, I think we did. [0:41] >> I think we're working for you now. [0:43] I think we're working for you now. [0:45] >> Did you guys get shares in OpenAI when [0:47] you sold this thing? [0:48] >> Yeah. [0:49] >> Yeah. [0:49] >> Everybody on the team. [0:50] >> So, we're all we're all on [0:51] >> We're on the same team. [0:52] >> Okay. [0:53] >> Yeah. [0:53] >> I How much money did you make? [0:56] >> Yeah, that's the [0:57] >> I I mean we got we got to turn the [0:58] tables a little bit. [1:00] Let's talk SpaceX. [1:02] >> Well, well, well, maybe before that we [1:04] can rewind a little bit cuz I do [1:05] remember the last time we were hanging [1:07] out in person. Uh I it was at [1:10] Katzenberg's event, right? You were [1:11] basically it was kind of this [1:13] interesting moment cuz in some ways we [1:15] were going through a mini correction, [1:18] right? Like chatbots, you know, grew [1:20] incredibly quickly. Agents were just [1:23] starting to work, [1:25] uh and it's been interesting to see how [1:27] the market did go through this [1:29] correction in Q4, but then realize hey [1:32] wow, agents are a thing and you had [1:34] pretty much perfectly called that in the [1:37] conversation that uh that we were having [1:40] which um [1:41] yeah, in some ways it's just been [1:43] uh it's been such a wild year for so [1:45] many reasons, but I feel like you had a [1:47] somewhat of a crystal ball back then. [1:49] >> Yeah. [1:50] >> Well, I mean [1:51] thank you. And I recall that [1:53] conversation and and the truth The fact [1:55] is we've gone through several mini [1:57] corrections over the course of last two [1:58] and a half years, right? There's been a [2:00] wall of worry. I mean, on my pad [2:03] podcast with Bill, [2:05] Bill Gurley, you know, we debated this, [2:07] you know, Bill was saying, will the [2:08] revenue show up? Will there be gross [2:10] margins? Will there be ROI? We're [2:12] overbuilding. Are we, you know, every [2:14] supply [2:15] constraint turns into a glut, and we saw [2:18] all that wall of worry last year. I [2:20] mean, I think for me, the turning point [2:22] was [2:23] when when we hit inference time [2:25] reasoning, and we really had this whole [2:27] other vector of scaling intelligence. [2:30] And I remembered having Jensen on the [2:31] podcast, he said, Brad, inference isn't [2:34] going to a 100x, not a 1000x, it's going [2:36] to 1 billion x because agents are going [2:39] to be talking to agents, right? And so, [2:42] I got very pilled. And then when we saw [2:45] Opus 4.6 in the beginning of December, [2:47] it was clear that we had crossed a [2:49] threshold of intelligence that offered a [2:51] level of utility that was fundamentally [2:53] different. [2:53] >> Yeah. [2:53] >> And if you were paying attention early [2:55] in December, you could see that coming. [2:57] But we started the year [2:59] with the market very skeptical as to [3:02] whether or not AI revenues would show [3:03] up. And let me tell you this, [3:05] had Anthropic not delivered its revenues [3:08] that it's delivered this year, I think [3:09] the stock market would be down 10 or [3:10] 15%. I think it's that important to the [3:14] entire narrative. Because the fact of [3:16] the matter is, OpenAI has not blown away [3:17] their numbers. Google has not blown away [3:19] their numbers. Like numbers have been [3:20] good, but the fundamental driver of of [3:24] of outperformance in terms of off-take [3:27] of AI revenues has been Anthropic, which [3:29] is the fastest-growing company in the [3:31] history of capitalism. So that buoyed [3:33] the entire AI segment. And it was when [3:35] they started posting those numbers, and [3:39] then they said on top of that, we're [3:40] doing it at high gross margins in a way [3:42] that in Q2 may in fact actually lead to [3:45] free cash [3:46] some positive free cash flow, the market [3:48] really ascended. Or remember, 2 months [3:50] ago, the market was basically down on [3:52] the year. [3:52] >> Yeah. [3:53] >> And a lot of these returns, we just had [3:54] two of the biggest months in the history [3:56] of Altimeter's public funds. That's 18 [3:58] years. I mean, we're going back a long [4:00] time. [4:01] But that's, you know, listen, I think we [4:02] picked some pretty good stocks, memory, [4:04] logic, etc. But I also think it's just a [4:06] function of the market delivered. These [4:09] companies delivered. You saw Dell's, I [4:11] mean, listen, Michael Dell, one of my [4:13] best buds, and you watch the the act, [4:16] you know, that he's delivering with [4:18] Dell. They just had AI server revenues [4:20] up 750% [4:22] year-over-year. Went from a $1 billion [4:24] business to a $16 billion business. This [4:26] stuff is real, but in order for it to [4:28] stay real, [4:30] we have to continue to see usage by [4:33] consumers that they're willing to pay [4:34] for, and growth in the enterprise small, [4:37] medium, and large that they're willing [4:38] to pay for, and growth in the sovereign [4:40] domain. I think it will occur, but [4:42] oftentimes, you know, there'll be some [4:43] pockets along the way here where, you [4:46] know, where where revenues won't be as [4:47] strong as people think. We'll have some [4:49] pullbacks. We could have 10 to 20% [4:51] pullbacks in the semiconductor stock as [4:54] just like run-of-the-mill consolidation. [4:56] >> Yeah. Yeah. [4:57] >> Right? Run-of-the-mill consolidation. I [4:59] mean, Micron has gone from a couple [5:01] hundred bucks to a thousand bucks. Dell [5:03] this time last year was, I think, 80 or [5:05] 90 dollars. It's now at $400. These are [5:08] seismic moves. [5:09] >> Yeah. Yeah. [5:10] >> Right? And so, uh yeah, I I fortunately [5:13] we were pretty bullish when other people [5:15] were skeptical. Um but [5:17] >> Yeah, it feels like it feels like [5:18] there's uh there's like this natural [5:20] reaction anytime there's good news, [5:22] someone has to dig up something that's [5:23] like a little bit bearish. Right now [5:25] we're seeing, you know, incredible [5:27] Anthropic revenues, and then there's the [5:28] questions about ROI on token maxing, and [5:30] how much is going on there. How are you [5:32] processing that? Are you thinking that [5:34] uh we'll see CEOs and management teams [5:36] on the next earning cycle uh sort of [5:39] start to dig into those numbers, or is [5:41] it just uh the models will get better [5:43] and they'll get better? [5:44] >> psychosis. [5:45] >> Yeah, well, that's the most extreme [5:46] version. The other one is yeah, we [5:47] actually did spend half a billion [5:49] dollars in a month and you know, a [5:51] quarter billion was super effective. So [5:53] that's what we're doubling down on in [5:55] the coming quarter. But how do you think [5:57] that shakes out? [5:58] >> Well, I mean, if I if I size up the [5:59] debate in Silicon Valley. [6:01] >> Yeah. [6:01] >> Right? There are the the bears who've [6:03] been bearish on AI for let's call it a [6:05] while. [6:06] >> And anything that comes out, anything [6:08] that comes out is actually just [6:10] >> And and so now [6:11] right? And so now they're saying, "Oh, [6:13] all this AI revenue is bullshit." First [6:15] they were saying it won't show up at [6:17] all. And then it showed up and then [6:18] they're saying, "Oh, it's all [ __ ] [6:20] because it's all token maxing and [6:21] there's no no ROI." So that's one side. [6:24] On the other side of the people who are [6:25] super AI pilled and they're like, "Oh, [6:27] no, this is perfectly, you know, Pareto [6:29] optimal. Everybody's spending the exact [6:30] right amount of money on tokens." which [6:32] we also know is not true. And the truth [6:34] lies in the center, okay? [6:36] When you have millions of independent [6:38] actors all making self-rationalizing [6:41] decisions like Altimeter on buying [6:43] tokens, right? I don't like I don't like [6:46] to waste money. [6:47] >> Yeah. [6:47] >> I'm spending money because I'm getting a [6:49] return. Now, will we will we experiment [6:52] with some things that don't provide a [6:53] return? Of course. I actually sent you [6:55] guys a slide on this. I think it's [6:56] pretty fascinating. I don't know if your [6:58] team can pull up, but this is [6:59] independent research that we did on this [7:01] question of of token maxing. And what we [7:04] did is we went to 300 enterprises, [7:07] right? And we just asked them, "Are you [7:10] starting to optimize your spend? And if [7:13] so, how much do you expect that you're [7:15] going to spend year over year over the [7:17] course of the next 12 months?" And if [7:19] you leave that chart up, what you will [7:21] see, guys, is that in the first [7:24] category, these are all people who are [7:25] actively optimizing. But they still [7:28] expect to grow revenues at over 50% over [7:31] the next 12 months. The second category [7:33] are people who are planning to optimize. [7:35] And even if they're planning to [7:36] optimize, they say they're going to grow [7:38] revenue at 90% and so on. So here's my [7:41] point on this, right? And this is across [7:43] 300 firms who use a multiple [7:47] of AI solutions. This is what they [7:48] expect of their AI [7:51] you know, API token usage. [7:53] >> Yeah. [7:54] >> So, what do we what what does that tell [7:55] us? [7:56] It tells us that of course people [7:58] optimize along the way. But we are so [8:00] early in the adoption curve. Right? [8:03] They're barely using coding today. [8:05] They're just getting on the coding train [8:07] and they haven't really even started on [8:09] using AI for knowledge work more [8:11] generally. [8:12] >> Yeah. [8:12] >> So, we're low in the use the penetration [8:14] of coding as a use case. We're almost [8:17] nowhere in in the penetration of [8:19] knowledge work more generally as a use [8:21] case. And then remember this, there are [8:23] very few enterprises you know, globally [8:25] that are even using AI. So, we're really [8:27] early in the curve of the people who are [8:29] actually using AI. So, I'm somewhere in [8:31] the middle. Of course, I believe that [8:33] optimization will continue. [8:34] >> Sure. [8:35] >> But my hunch is that Anthropic and [8:38] OpenAI, these companies will continue to [8:40] grow right through the optimization [8:42] because the growth curve on penetration [8:45] of both enterprise and use case is so [8:47] steep. But you know, we'll see. [8:50] >> Yeah. [8:50] Is this is this a zero-sum market where [8:54] every dollar spent on tokens comes out [8:57] of a SaaS company? How are you [8:58] reflecting on the SaaS apocalypse [9:00] because we all saw what happened in the [9:02] market, but there's been some really [9:04] good news lately. How have you processed [9:06] that? [9:07] >> Well, I I I popped on CNBC for a second [9:10] yesterday and was talking about [9:11] Snowflake as an example and the stock [9:13] was up what 35% yesterday. Now, of [9:15] course, just to be fair, it's only up [9:17] 10% for the year compared to a company [9:20] like Micron up 200% for the year or a [9:22] company like ARM up 200% for the year. [9:25] But they did bounce back and what what I [9:27] think we're starting to see is the [9:29] bifurcation. [9:30] >> Mhm. [9:30] >> There are companies that are in the [9:32] token flow. So, all these software [9:33] companies we just lumped together. [9:35] >> Yeah. [9:35] >> Right? We treat them as though they're [9:36] all equal. But there's certain software [9:38] companies, Databricks, Snowflake, and [9:40] ClickHouse, all of which were investors [9:42] in, which is very clear to me, they're [9:44] in the token flow. As you consume more [9:46] tokens, the amount of your database [9:48] queries goes up. I see it at Altimeter, [9:51] right? In fact, our database queries are [9:52] growing faster than our token usage, to [9:55] give you a sense. And this is I think [9:56] how So, now they've proven they're in [9:58] the token flow, so they're starting to [10:00] get some love from an AI multiple [10:02] perspective. That's very different than [10:04] a company I think like like Salesforce. [10:06] And I love Marc Benioff and and got you [10:08] know, he if anybody can you know, get in [10:10] the token flow, it'll be him. But the [10:12] reality is [10:14] the the front-facing solutions that they [10:16] offer are more competitive with the [10:18] models than something like Snowflake. [10:20] Snow- Snowflake's the enabler of the [10:22] models, whereas I think that Salesforce [10:24] competes a bit more. So, it's going to [10:25] be more challenging. But I also I I've [10:28] heard so much about about this [10:29] SaaS-pocalypse. And And listen, I did a [10:31] pod with Saqja, I don't know, 18 months [10:34] ago where he caused a stir by saying [10:36] software's a thin user interface on top [10:38] of a crud database, and Benioff and [10:40] everybody freaked out. They're like, [10:41] "What are you saying? It's way more than [10:43] that, right?" And then Bill and I did a [10:45] pod, is software dead? So, it's not like [10:47] this is new. But then everybody started [10:49] freaking out in December. All these uh [10:52] multiples reset. But the question is, [10:53] what did they reset to? [10:56] Okay? And this is what I want to focus [10:57] on here. So, if you show this slide that [11:00] that I prepared for you fine esteemed [11:02] gentlemen, [11:04] you know, what the slide shows is that [11:06] the multiple correction just took [11:08] software from a place where they were [11:10] way more expensive than the market [11:12] multiple [11:13] and brought them into the category of [11:16] the market multiple. Right? So, now [11:18] they're trading at about 22, 23 times [11:22] real SBC included GAAP earnings. [11:25] >> Yep. [11:25] >> That's about where the market is [11:27] trading. So, now just follow me on this. [11:30] Software is trading mostly software [11:32] names are trading at a higher multiple [11:35] than Nvidia. [11:37] >> Yeah. [11:37] >> Right? Nvidia's trading about 13 times [11:40] earning for 70% growth for the thing [11:42] that is the most essential thing in AI, [11:47] and they're they're they're at twice the [11:49] multiple. So, like when I hear everybody [11:50] crying that hey, these multiples aren't [11:52] fair, it looks to me like the multiples [11:54] reset from an above market multiple [11:57] where everybody thought this the [11:58] software revenues and and earnings were [12:00] impenetrable to now they're saying, [12:02] well, I don't know. Some of this may be [12:04] three, four, five years out will be [12:05] replaced. So, we're going to raise the [12:07] discount rate, we're going to lower the [12:08] multiple. They've only lowered it to the [12:10] market multiple. [12:11] >> Yeah. [12:12] >> Let me just suggest that there's a [12:13] possibility these trade well below the [12:15] market multiple. [12:16] >> Sure. [12:16] >> Right? I'm not I'm not wishing for that, [12:19] but I'm just saying there's a [12:20] distribution of potential outcomes here. [12:23] If you get on the AI train, if you get [12:25] in the token flow, you're going to get a [12:27] above market multiple. If you don't, [12:30] if you slow down and it looks like every [12:32] time the computational intelligence [12:34] improves, your business gets worse, then [12:37] I promise you they will trade below the [12:39] market multiple and there's and there's [12:40] more room to the downside. So, for us as [12:42] investors, you know, Warren Buffett has [12:44] this [12:45] you know, this old metaphor, you know, [12:47] there's the easy basket, there's the [12:48] hard basket, [12:50] uh [12:50] or the the yes basket, the no basket, [12:52] the two hard basket. [12:54] >> For me, software today is generally in [12:56] this in the two hard basket. [12:57] >> Yeah, it's it's notable cuz you've been [12:59] saying that I I think for months now. [13:02] >> And there's a lot of people now that [13:04] there's been a stabilization that that [13:06] say like, oh, I'm I'm smart enough. I [13:08] can I can I I'm I can outsmart the [13:10] market here. And and like you're saying, [13:12] even with where multiples are now, you [13:14] still could be catching a falling knife. [13:17] Um I wanted to ask you about uh the [13:21] uh [13:22] a potential data center moratorium [13:25] and how, you know, the likelihood of [13:28] something like that in your view, how [13:29] that would [13:30] if you have less capacity coming online, [13:34] that would obviously be bad for, you [13:37] know, [13:38] chip companies, various companies in the [13:40] hardware supply chain, but it could be [13:42] great for people that are actually have, [13:46] you know, basically like have tokens to [13:48] sell because they would potentially get [13:50] more pricing power. How do you [13:52] >> it's bad for everybody. It's bad for [13:54] everybody, but most importantly, it [13:55] would be horrific for America. [13:57] >> Yeah. [13:57] >> And lest we be overconfident in Silicon [13:59] Valley, let's remember the activists, a [14:02] small group of activists, shut down [14:05] supersonic technology, [14:07] and a small group of activists shut down [14:10] all nuclear clean energy in this [14:11] country. Okay, we have 100 fusion [14:14] reactors being built in China. We have [14:16] one in the United States. It's a [14:18] disaster that happened, and so we can't [14:21] take for granted that the kooks who are [14:23] calling for data center [14:25] a moratoriums, right? Which Just think [14:27] about this for a second. [14:29] All of our GDP growth is coming from the [14:32] fact that we are building data centers [14:33] and driving AI and driving productivity [14:36] improvements in the economy. A data [14:38] center moratorium would thrust us [14:40] straight into a recession and high [14:42] unemployment. Secondly, it would seed [14:45] the entire global game to China. [14:47] Like, overnight, we would lose to China [14:49] in the global AI race, which is not just [14:51] about AI, it's about economic security, [14:54] it's about jobs, and it's about national [14:56] security. [14:57] >> Yeah. [14:57] >> So, it literally is insane that we would [15:00] do this. I can't even believe there are [15:01] people talking about it. However, [15:04] what are Why are they talking about? [15:06] Because people are concerned. Local [15:08] communities are concerned. I just got [15:09] back from celebrating my mother's 90th [15:11] in rural Indiana, [15:13] um, you know, over the Memorial Day [15:15] weekend, and what I'll tell you there [15:16] >> Happy birthday. [15:17] >> Happy birthday. [15:18] >> Uh, she's incredible. She is going [15:20] incredible. But you think about a place [15:22] like Mishawaka, Indiana. [15:24] >> Yeah. [15:25] >> Where, you know, they're building a data [15:27] center. I mean, folks here [15:28] they they're worried about their jobs, [15:30] they're worried about their kids having [15:31] jobs, and then they're told by these [15:33] crazy activists who show up in their [15:35] town, they're not going to have any [15:36] water and their electricity bills are [15:38] going to go up. So, can we blame these [15:40] people for being a little agitated about [15:42] what's going on? So, I'm actually [15:44] working on an initiative I'm not [15:45] prepared to announce today, but with [15:47] like everybody in the value chain, all [15:49] of the cloud companies, all of the [15:51] Nvidias and AMDs and, you know, and [15:53] off-takers etc. and the White House that [15:55] would deliver a very tangible and [15:58] profound dividend to the communities [16:01] that we're building data [16:02] >> There we go. [16:03] >> And I think it's I think it's the [16:05] there's a very elegant solution there. [16:07] You're the guy to do it. You're the hero [16:09] that American capitalism deserves. You [16:11] got Trump accounts done. I feel like [16:13] this is a good good next act for you. [16:16] >> Um well, I'm I'm in the mix. Um I'm [16:19] happy to do my part. There are [16:20] extraordinary people around the table. [16:23] Um but here's the thing. We have to [16:25] build the socio-political bridge for the [16:27] next 3 years, right? In 3 years it's [16:29] going to be obvious, I think, the [16:31] abundance and the benefits that AI is [16:33] driving for us as consumers. Everybody's [16:35] going to have their own personal [16:36] assistant in their pocket, right? For [16:38] next to nothing. Think about that. [16:40] >> Yeah. [16:40] >> Can do your calendar, can order your [16:42] food, can, you know, get you a new black [16:44] t-shirt, send mom a birthday present, [16:46] all the things. And every enterprise is [16:48] going to have things that up-level us [16:49] all as humans. So, I am firmly, just [16:52] like uh you know, John Maynard Keynes [16:55] was at the start of the Industrial [16:56] Revolution, I am firmly in camp optimism [16:59] about technological progress. But I'm [17:01] also not head in the sand about the [17:03] disruption and the concern people have [17:05] for the next 3 years. So, we have to [17:07] give them tangible benefits that get us [17:09] over that bridge. I think we're going to [17:11] do it. I'm feeling pretty optimistic [17:13] about it. But you're right to bring it [17:14] up and you're right to be concerned [17:16] about it. We cannot take it for granted. [17:18] >> Yeah, the I mean this just goes back. I [17:21] think it's entirely fair that [17:22] individuals you know, if you say I'm [17:24] going to put an AI factory in your [17:26] backyard, okay, it's going to create [17:28] jobs [17:29] uh briefly and then, you know, some some [17:32] some maintenance. I think it's totally [17:33] fair for people to not want it in their [17:35] backyard because there's some they they [17:38] they perceive some risk uh and uh [17:41] there's no direct benefits cuz they can [17:42] just get AI anywhere, right? It doesn't [17:44] matter [17:45] >> to we're going to we're [17:46] >> But there's a solution. [17:47] >> Yep. [17:48] >> Um how are you thinking about adoption [17:52] curves? Uh it feels like part part of [17:54] the reason that we've had, you know, [17:56] these kind of like rolling uh [17:58] corrections is that, you know, [18:00] technology gets adopted really quickly. [18:02] People assume that it's just a a [18:03] straight line forever, but then there's [18:05] a new capability, a new a new technology [18:07] and it feels like stuff is just breaking [18:09] through like instantly. [18:11] Um where have you like I are you [18:14] adopting new frameworks internally to to [18:17] try to understand how quickly new [18:19] products can uh get to market? [18:21] Obviously, enterprise is is different, [18:24] but uh it feels like the line between [18:26] consumer and enterprise, at least in, [18:28] you know, coding has never been more [18:30] blurred. [18:31] >> For sure. I mean, listen, I I I think [18:34] about when I got into the game, guys, [18:35] 1999, 2000, and we had about 35 million [18:39] people connected to broadband internet, [18:42] right? We all saw what Amazon was going [18:43] to be, but where we got over our skis, [18:46] right, is we thought it would come a lot [18:47] faster and we forgot that there were [18:48] only 35 million people connected to [18:50] broadband internet. [18:51] >> Yeah. [18:52] >> Today, we have 4 billion, three, four [18:54] billion. Like, the rate of diffusion [18:56] and the the the the magnitude of [18:58] diffusion is radically, radically [19:00] different. And think about this, [19:02] we have a natural constraint on how fast [19:05] we can go because we only have so many [19:07] memory wafers in the world. [19:09] We only have so many logic wafers in the [19:11] world. We only have so much powered [19:13] shell in the world. That means we can [19:15] only produce so many tokens. [19:17] Okay? And it's almost as though in [19:19] 1999-2000, we could only lay so much [19:22] fiber. I've said this a thousand times. [19:25] When we were putting down the fiber in [19:26] 2000, we called it dark fiber for a [19:29] reason. There was nobody using it, and [19:31] we knew there was nobody using it when [19:33] we put it in the ground. There's not a [19:35] dark GPU in the world today. [19:37] >> Yeah. [19:37] >> Okay? There's not a dark token in the [19:39] world today. So, I think it's a very [19:41] different thing. I think it's a healthy [19:42] thing. We have this wall of worry. We [19:44] can't build that much supply, and I [19:46] would say if I look at every company, [19:48] what did they report on their earnings [19:49] calls? Google was token constrained. [19:52] They said, "If we had more tokens, we'd [19:53] be able to generate more revenue." Same [19:55] for Amazon. Same for Microsoft. Same [19:57] from Open AI. Same from Anthropic. The [19:59] world demands more intelligence. [20:02] Intelligence can only be produced with [20:04] tokens, and we have physical limits to [20:06] the amount of tokens we're going to be [20:07] able to bring online. So, yes, we will [20:09] have these waves, but I think the rate, [20:12] the parabolic rate at which these new [20:14] models are going to produce [20:16] intelligence, I think we're going to be [20:17] blown away over the course of the next 9 [20:18] months. You talk independently, uh you [20:21] know, to Michael Truelove and the guys [20:22] at Cursor, and now, you know, taking [20:25] over x.ai, or you talk to uh the guys at [20:28] Anthropic or Open AI, and they kind of [20:29] look you in the eyes with that [20:31] Oppenheimer look, and they're like, [20:33] "We're here." [20:33] >> Yeah. [20:34] >> Right? Like we're like like We are going [20:37] to Think about this. [20:38] Open AI and Anthropic combined to start [20:40] the year had 3 gigawatts of compute. [20:42] Three, combined. [20:43] >> Mhm. [20:44] >> They're going to end the year closer to [20:45] 10, and end next year closer to 20. [20:48] >> Yeah. [20:49] >> We're making algorithmic improvements. [20:51] We're making massive steps up the [20:53] scaling law because the amount of [20:54] compute we're going to have available to [20:56] us. Think about, you know, macro harder [20:58] and macro harder that Cursor's going to [21:00] now be able to train a frontier-level [21:02] model on. So, we've got incredible [21:04] competition in America. We got the right [21:05] amount of compute coming along. I don't [21:07] worry about the bubble as much even [21:09] though I know that, you know, [21:11] there there will invariably be, you [21:13] know, some months that revenue doesn't [21:15] grow as fast. [21:16] I'm really worried about making sure [21:18] that America stays foot on the [21:21] accelerator competing globally and [21:23] winning the AI race. Like this is going [21:25] to lead to a moment of of abundance for [21:28] our economy and it's only through great [21:30] national wealth that we can raise the [21:32] floor for everybody else. [21:33] >> Yeah. No, that makes [21:35] >> total sense. [21:35] >> Uh there were some reporting this week [21:37] that Meta is hiring FTEs. [21:41] I was sort of surprised to see them [21:43] going into the enterprise because it [21:45] feels like they have every advantage on, [21:49] you know, consumer. They have the [21:50] billions of users. They have [21:53] uh they have, you know, exciting [21:55] hardware. [21:56] All all these things. [21:58] Um [21:59] how much did you Was that Was that [22:00] surprising to you at all? Do you expect [22:03] more companies that weren't [22:04] traditionally, you know, [22:06] enterprise-focused to say, "Hey, there's [22:08] tens, maybe hundreds of billions of [22:09] dollars of revenue here. We should be We [22:12] should be in this market." [22:13] >> I mean, the second you start spending a [22:15] hundred billion dollars on capex [22:16] annually, [22:18] okay, you run into the AWS problem. [22:21] What's the AWS problem? Now I have all [22:23] this compute but I don't use it every [22:25] day equally. [22:26] >> Yep. [22:26] >> Right? Jeff built AWS because he said, [22:29] "I have to build my capacity for [22:31] Christmas Day or that, you know, the [22:33] week leading up to Christmas." [22:34] >> Black Friday. [22:36] But he's like, "The rest of the year, [22:37] half of that stuff sitting idle, it's [22:38] expensive as hell. So, I may as well [22:40] rent it to everybody else, right? Turned [22:42] into [22:43] you know, a Blockbuster business but it [22:45] made his core business better [22:46] >> Mhm. [22:47] >> because he could he could build to Black [22:49] Friday, right? And nobody else could [22:51] because they didn't have AWS. So, that's [22:53] why Elon has launched EWS, right? Elon [22:55] Web Services. [22:57] You know, with his compute and he's, you [22:59] know, signed up a big first customer [23:00] with Anthropic. Listen, nobody on Earth [23:03] is better at turning electrons into [23:04] tokens than Elon, right? So, expect a [23:07] lot more data centers out of Elon, [23:08] expect them on Earth, and eventually in [23:10] space. And I think that changed the [23:13] whole tenor of the SpaceX IPO. Both the [23:15] Corsair deal and the Anthropic deal, I [23:17] think that went from, you know, people [23:19] being slightly concerned about it to [23:20] people being quite excited about it. I'm [23:22] happy to, you know, to to unpack that. [23:25] So, I think that for Meta, [23:28] if they're going to be in the game and [23:30] spending that much money, listen, Susan [23:32] Li is, you know, incredible over there [23:35] as the CFO. [23:36] And I'm sure they're looking at the [23:38] strategic plan and Mark is saying, I [23:40] want to build even more because that guy [23:42] is never going to give up the race, [23:44] right? To frontier level AI. None of [23:46] these guys want to give up that race. [23:48] And so, they just have to figure out [23:49] ways to monetize everything that they're [23:52] building. Do I worry as a shareholder at [23:54] some level that it's, you know, that's [23:56] hard. That's hard to take a business [23:57] that's been 120% consumer and say, [24:01] "Okay, now we're going to be in the [24:02] business of AWS and maybe even in the [24:04] business of enterprise level agents." I [24:06] think it is hard. I think they're up for [24:08] the call. And remember, you suggested [24:11] the merger between, you know, [24:13] product-led growth, these coding agents [24:15] kind of feel like consumer adoption. [24:17] >> Yeah. [24:17] >> So, there's a lot of shared consumer DNA [24:20] with what's going on in the enterprise [24:21] today. So, they may surprise some folks, [24:24] um but um [24:25] >> And Meta does have links into like [24:28] hundreds of thousands of businesses [24:29] through the ads platform. So, it's not [24:31] like they don't have any relationship to [24:32] businesses. They do, so. [24:34] >> One more that I I was curious to get [24:36] your thoughts on Kirkland and Ellis uh [24:39] is talking about investing half a [24:41] billion dollars into their own [24:43] uh software to help run their firm. Uh a [24:46] lot of people [24:48] pushing back on that. Historically, you [24:50] take a a firm that doesn't have strong, [24:52] you know, software competency and they [24:54] spend, you know, hundreds of millions of [24:56] dollars on their own software. There's a [24:57] lot of examples where that hasn't gone [24:59] well. Yet at the same time, [25:01] software making software today is wildly [25:04] uh different. And uh it's very possible [25:08] that that things are changing, [25:10] especially if you can get the right [25:11] partners around it. I know they have [25:13] some great partners. Uh do you expect [25:15] more companies of that scale, services [25:17] businesses to want to try to own as much [25:20] of the stack as possible and not be [25:22] reliant on, you know, the Harveys or the [25:25] Lagores of the world? [25:26] >> I mean, what else are they going to do? [25:27] I mean, it's kind of like what else are [25:28] you going to announce? Oh, just we give [25:30] up. You know, so like and like they got [25:33] to do something. The competition is [25:35] coming straight at them. [25:36] >> Yeah. [25:37] >> I I don't think it's a high probability [25:39] bet, personally. Like if I was a partner [25:41] at Kirkland and Ellis and somebody [25:42] pitched me on that, I'd say, "Ah, I I I [25:44] I'm I'm not sure that's the highest and [25:45] best outcome here." Um so, what is an [25:47] alternative outcome? Well, [25:49] good friend, you know, Josh Kushner, [25:51] what he's doing at Thrive Holdings, [25:53] right? Where he's buying accounting [25:54] companies. And now I have somebody who's [25:56] just like deep in the weeds recruiting [25:58] the best engineers in the world, deep [26:00] partnership with OpenAI. I saw Greg [26:02] Brockman retweet the great the work that [26:04] they're [26:05] that they're benefiting all these [26:07] accounting firms. Like they're driving [26:08] just huge productivity gains in these [26:10] accounting firms. So, it seems to me [26:12] that that's a more likely outcome, you [26:14] know, uh [26:15] a Thrive Holdings buying a Kirkland and [26:17] Ellis and saying, "Now we're going to, [26:19] you know, take this thing and AI [26:21] turbocharge it." I think you're going to [26:23] see a lot of that out of private equity [26:25] firms, out of out of firms like Thrive [26:27] Holdings. I think you're going to see [26:28] take privates where people do that on an [26:30] individual company [26:32] basis. But am I confident that software [26:35] has gotten so easy that a law firm that [26:37] gets up every day and thinks about [26:38] writing legal briefs is all of a sudden [26:40] going to write killer legal software to [26:42] compete with OpenAI and Anthropic? Uh I [26:44] think that's unlikely. [26:46] >> That's hard. [26:47] >> Uh what what is your thinking around the [26:50] series A, B, C, these, you know, earlier [26:55] growth rounds, feels like a lot of [26:57] investors are just kind of frozen. They, [27:00] you know, you were talking earlier about [27:02] not necessarily frozen in terms of their [27:04] activity. They're doing a lot of deals, [27:06] but they maybe don't have as much [27:08] confidence knowing [27:10] >> What will get steamrolled [27:11] >> what will get steamrolled in in the [27:13] future. You were talking about being in [27:15] the token flow. Is that like where you [27:17] feel comfortable deploying at this early [27:20] stage where you, you know, you're [27:21] betting on a, you know, a 10-year [27:23] outcome? [27:24] >> Yeah, I mean, listen, I I think we all [27:26] have to have the humility in these [27:28] moments to know that looking out 10 [27:29] years is almost impossible. [27:31] >> Mhm. [27:31] >> Um, looking out 10 months is pretty damn [27:33] hard. Um, but I would say if you just [27:35] looked across our portfolio and, you [27:37] know, I think Altimeter is performing [27:39] better than it has anytime in its [27:40] 18-year history. You know, our [27:43] early-stage team, I think [27:44] >> Overnight success. [27:47] >> Awesome, you know, awesome work uh, on [27:50] the early stage side, but if you look at [27:51] the type of stuff that we're investing [27:54] in, it is in the token flow. [27:55] >> Mhm. [27:56] >> Right? We're building to those compute [27:57] shortages. You know, we had the Cerebras [27:59] IPO, you know, last week. We had been in [28:01] that for 9 years. [28:03] Um, you know, investors, you know, Grok, [28:06] we're looking at a lot of other [28:07] semiconductor type businesses. We're [28:09] looking at a lot of compute data center [28:12] type businesses. Um, and you know, I I [28:15] you just had you know, your prior guest, [28:18] you guys were talking about all the [28:19] stuff you're doing in military [28:20] modernization and the stuff that's [28:22] adjacent to AI but are benefiting from [28:24] AI. We're doing a bunch of stuff there, [28:26] um, in modernization of the military. [28:28] So, I think you find places that are [28:30] either in the token flow or benefiting [28:32] from the token flow. [28:33] >> Mhm. [28:33] >> And then I would say in growth, like [28:35] we're just not doing a lot in what I [28:36] would call inflection stage growth. This [28:39] is the companies at 5, 10, 15 billion. [28:42] You know, we've really made massive [28:44] bets, the biggest bets in the history of [28:46] Altimeter between open AI and Anthropic, [28:48] which uh, you know, is consuming [28:50] billions and billions of dollars. [28:52] And so we think they are the principal [28:54] beneficiaries. And then on the public [28:55] side, [28:57] for 3 years now, we've basically had [28:58] 100% of the portfolio in AI and compute. [29:01] And you know, as I sit here today, even [29:03] though it's come up a lot, you know, [29:05] Hynix is still trading at a single digit [29:06] multiple and Micron's trading at a [29:08] single digit multiple and Nvidia's [29:10] trading at 13, you know, times. And you [29:12] say, how is that even possible? Nvidia's [29:14] up 15X, like better than a venture [29:16] market return over 3 years. Like Like [29:19] think about that. Like all the venture [29:20] returns have been had in the public [29:22] markets, by the way, guys. [29:23] >> earnings have come. [29:24] >> Yeah. [29:24] But but but their multiples have come [29:26] down. [29:26] >> Yeah, cuz the earnings have come. [29:28] >> have actually come down. This is the [29:29] cheapest multiple Nvidia has traded at [29:32] in a decade right now. [29:33] >> Crazy. [29:34] >> Okay? And by the way, I think their [29:36] growth is going to continue to sustain. [29:38] They're now taking 50% of their free [29:40] cash flow and returning it by way of [29:42] dividend or buying back stock. I would [29:44] encourage Jensen to do 70 or 75%. I [29:48] think if he does that, by the way, a [29:49] prediction, [29:50] you know, look who invested in Apple the [29:53] second they bought back or the second [29:55] they committed to 50% 75% of their free [29:58] cash flow returning to investors, Warren [30:00] Buffett. [30:01] >> Yeah. [30:01] >> One of the greatest investment returns [30:03] in history, right? And so once you make [30:05] that cross that threshold, and I think [30:07] this is, you know, they're running that [30:08] business incredibly well. So the public [30:10] markets we've had, you know, basically [30:12] 100% in AI AI and compute, we're [30:14] basically there, [30:16] you know, today. So I think it is [30:18] harder. [30:19] If you're a Series B or Series C [30:21] company, think about what we used to do [30:22] in software. If it's Series A, you had a [30:24] couple million in revenue, and then [30:26] Series B, I don't know, you had 20 [30:27] million in revenue. [ __ ] you would have [30:29] a line out the door of people who wanted [30:31] to do that deal. There's a you wouldn't [30:33] have a single taker today. [30:34] >> Mhm. [30:35] >> Not a single taker. [30:36] >> You mentioned [30:38] something I think that resonates with [30:39] everyone. It's very hard to predict [30:40] what's going to happen in 10 years. [30:42] Obviously, your job is to, you know, [30:45] look at the global trends and names. [30:46] But, I want to know about the Trump [30:49] accounts and I want to know about [30:51] investing for the next generation, for [30:53] children, advice. Also, get me up to [30:56] speed on the program, what's rolling [30:58] out, what's the progress, but then what [31:00] is advice to parents in an uncertain [31:03] time where setting their children up for [31:05] success is maybe more critical than [31:08] ever. [31:09] >> Well, um [31:11] the update is that after 4 years of [31:15] working on this and, you know, getting [31:17] it passed into law last July 4th, the [31:19] Invest in America Act as part of the the [31:21] big beautiful bill, [31:23] you know, it's set to [31:25] launch and be funded on this July 4th. [31:28] >> That's great. [31:28] >> But, we launched the app, guys, [31:30] yesterday. So, you can download the app. [31:32] Every single family, you should tell [31:34] every family you know who has a kid [31:36] >> Yep. [31:36] >> they should download the app for their [31:37] kids, get their kids signed up. There [31:39] are 35 million kids in America under the [31:41] age of 10, [31:42] okay, who get at least 250 bucks. [31:45] >> Yeah. [31:46] >> So, if you're basically born after [31:47] January 1st, 2025, so think about like [31:49] under two, you get 1,000 bucks in the [31:51] S&P 500. [31:53] >> Yeah. [31:53] >> If you're between two and 10, [31:55] you get 250 bucks. Most of those kids [31:57] will get 250 bucks from Michael and [31:58] Susan Dell. If you live in Indiana, [32:00] you'll get an extra 250 from me. If you [32:02] live in Connecticut, you'll get an extra [32:04] 250 from Ray Dalio. If you live in [32:06] Oklahoma, you'll get 250 from the state [32:07] of Oklahoma, okay? And that's just for [32:10] starters. We have thousands and [32:11] thousands of companies that are [32:12] >> of billionaires in states that you [32:14] didn't name and they're starting to get [32:15] on the list. [32:16] >> listen, listen. They've heard from you, [32:17] I'm I'm sure. If they haven't heard from [32:19] you, they're going to. It's It's coming. [32:22] And by the way, the generosity, this is [32:24] the Giving Pledge 2.0. [32:26] >> Yeah. [32:26] >> We have trillions and trillions of [32:28] dollars that are going to change hands [32:29] in this country. This is the single most [32:31] efficient way for somebody like me to [32:34] fund the next generation. 100 cents on [32:36] the dollar goes to the kid. It compounds [32:38] for 18 years for their lifetime. It [32:40] makes them a capitalist and owner. We [32:42] know they're more likely to graduate [32:43] from high school and college, more [32:45] likely to start a business, more likely [32:46] to to buy a home. The societal ROI on [32:49] this is off the charts. So, we launched [32:51] it yesterday. Get a Get a rip of this, [32:53] man. It is now the number three app in [32:56] the United States. [32:58] The number three app. We just passed [33:00] Google. [33:01] We're only behind ChatGPT and Claude. [33:04] It's incredible. [33:05] >> You're other You're coming for your [33:07] kids. [33:09] >> It's in all three of the top app store [33:11] apps right now. [33:12] >> Total Gusner total Gusner victory. [33:15] >> It's a total Gusner victory. [33:16] >> Well, well, I would say, [33:18] um, you know, kudos to Vlad and the guys [33:20] at Robinhood and BNY and Joe Gabbia at [33:23] the National Design Studio and frankly, [33:26] the whole team at the Treasury [33:27] Department led by Luke Petitt and the [33:29] Treasury Secretary. [33:31] This is the way government should be [33:32] done. A citizen had an idea. He was able [33:35] to go to Washington and actually get a [33:37] law passed. And then we put together a [33:40] SWAT team of people who are experienced [33:42] building these things to build them. And [33:44] then the consumers, i.e. the citizens of [33:47] America who pay for this [ __ ] said, [33:49] "Hey, we love that thing." And and and [33:51] and bid it up on the app store. So, we [33:53] have a lot of people downloading the [33:55] apps. There are a lot of improvements [33:56] coming. So, be patient with us, but [33:59] download the app. Get you Get your kid [34:01] on the path to compounding. Um, on July [34:04] 4th, guys, the money turns on. So, [34:06] you'll Every parent's going to see that [34:08] their kid owns a little Nvidia, a little [34:10] Microsoft, a little Walmart, right? [34:12] Their little slice of all the top 500 [34:14] companies in America. And on July 6th, I [34:16] hope that we have a joint bell ringing [34:18] of the New York Stock Exchange and the [34:20] Nasdaq from the Oval Office, uh, to [34:22] really signify the start of of of the [34:24] trading these accounts. Of course, [34:25] parents don't have to know anything [34:27] about investing. It all goes into the [34:28] S&P 500. [34:30] >> Yeah. [34:30] >> Okay? But I'm cajoling some of our [34:32] friends. [34:34] You would know their names. [34:35] I think it would be amazing if we had [34:38] some of our friends gift a share of the [34:40] most amazing companies in America. You [34:42] know, the Facebooks, the SpaceXs, the [34:45] OpenAIs. How about if they all gave just [34:47] a share of those companies to every kid [34:48] in America? [34:49] >> Yeah, be incredible. [34:50] >> We we are going to change and reorient [34:52] how the 70% of people who have felt left [34:55] out and left behind, they are not owners [34:58] of capital, okay? We need to get them on [35:00] the compounding journey. They need to [35:02] feel like they're on team America, [35:03] they're in the game. This does that for [35:06] every child. This is not a 529 account [35:09] for the top 10% of Americans who can [35:12] afford to save. This is for everybody [35:14] and it's so gratifying. I was in Durham [35:17] last Friday. I adopted a school. There's [35:19] 700 kids. $250 to every one of the kids. [35:23] Now, a lot people say, "Well, how did [35:24] you do that?" Well, it's $250 * 700 [35:27] kids. They made a Google spreadsheet. [35:29] They got them all signed up. I give the [35:30] principal $150,000 and she QR codes the [35:33] money and teaches the accounts. [35:35] >> Mhm. [35:35] >> Okay? Everybody in America can adopt a [35:37] school. Raise a little bit of money, go [35:40] to your principal and say, "We want to [35:41] juice up these accounts for all the [35:42] kids, get all the kids signed up." And [35:44] the teachers there, this was a school [35:47] that's 75% black and Latino serving the [35:49] rural poor in Durham. The level of [35:52] excitement, a mom came up to me crying. [35:53] I never thought my kids would, you know, [35:56] own anything. The teachers so excited to [35:58] teach the kids about what it means to to [36:01] own something. You know, I grew up in [36:02] rural Indiana and we had zero. [36:04] And as I said to the president, when [36:07] you're at zero, it's a despondent place [36:09] to be. You don't know how to get to one. [36:11] >> Yeah. [36:12] >> The hardest move in the world is going [36:14] from zero to one. [36:16] One to two is easier and two to three is [36:17] easier yet. We're going to get all of [36:20] these kids from zero to one on this [36:22] compounding journey. If you start with a [36:23] thousand bucks and you save $50 a month, [36:26] it's $50,000 at age 18. There's no [36:29] reason we can't put every kid in America [36:31] on that journey. And to celebrate our [36:33] second 250 years, right? We're we're [36:36] launching a natural [36:38] you know we're going to launch this as a [36:39] dividend for every kid in America. So I [36:41] want to make sure that they all sign up [36:42] starting in 2027, [36:45] the 3.7 million kids born in 2027, it [36:48] will be automatic. [36:50] >> Yeah. [36:50] >> Get your social security number, you get [36:52] a Trump account. [36:54] >> Um [36:54] >> And then we just need to get every small [36:56] We're giving money to the We have 80 [36:58] kids uh you know to our you know roughly [37:02] 35 employees, they're all going to get [37:04] 500 bucks at the end of the year into [37:05] their Trump accounts. I'm just going to [37:07] QR the money [37:08] by my team into their accounts. You guys [37:10] should do it for all the companies [37:11] you're involved in. [37:12] >> Yeah, that's great. [37:13] >> And and really spread the word. Small, [37:15] medium, large business, realtors, [37:17] restaurants, everybody can do this. And [37:20] so we've created an open-source platform [37:23] of universal private ownership. [37:26] >> Yeah. [37:26] >> Where the families have the title and [37:28] they have the dignity the dignity of [37:31] savings. A 401k for life for every [37:34] single American citizen, I think it's a [37:35] game-changer for the country. [37:37] >> Yeah. [37:37] >> You did it. [37:38] >> You did it. It was fantastic and [37:39] >> did it. Oh [ __ ] I forgot. [37:42] >> No, I remember you you you you talking [37:44] about this and [37:46] uh and you know as as as much as respect [37:49] as I have for you, [37:51] uh I put it in the in the it's too hard [37:53] bucket, you know? I I I put it in the it [37:56] like this is a thing that is just too [37:57] hard for anyone, even even the best. And [38:00] fortunately it was not, which is [38:01] fantastic to see as the ultimate white [38:03] pill. So thank you. [38:04] >> That's amazing. [38:05] >> It's a we're just you know it's still [38:06] day one, but you know we're off to a [38:08] good start here and [38:10] you know I think in the fullness of [38:11] time, as the president said, we estimate [38:14] over 15 years it could transfer three to [38:16] four trillion dollars of wealth from [38:18] people who have it to the people who [38:19] would otherwise have zero. [38:21] >> Yeah. [38:21] >> Um and um you know, the president has [38:24] said he thinks it's going to be his [38:25] biggest legacy. [38:26] >> Yeah. [38:26] >> Um, to me I think it'll be more [38:28] impactful in the fullness of time than [38:29] social security. Because the difference [38:31] is you actually own this. [38:34] >> Yeah. [38:34] >> You actually own it. It's not a [38:35] government program. [38:36] >> Yeah. [38:37] >> Yeah. [38:37] >> This is a private account, private [38:38] ownership that can compound through your [38:40] life. [38:40] >> And you and you have to imagine that you [38:42] know, if if you get to that place where [38:44] you know, there's there's a whole new [38:46] generation that's [38:47] you know, becoming an adult, starting a [38:49] family with 50, 100, 200,000 dollars. [38:52] That's a down payment on a house. All of [38:54] a sudden that can underwrite more [38:56] building of houses because there's more [38:57] buyers in the market. There's there's a [38:59] whole bunch of market forces that I [39:00] think will knock on from this in 20 [39:02] years that could be incredibly positive. [39:04] So, I'm I'm extremely excited about it. [39:07] >> Indeed. No no doubt about it. It's a [39:10] you know, you're going to you're going [39:11] to hear a lot of a lot out of us over [39:13] the course of the next several months. [39:14] But it's you know, listen, I also should [39:16] mention I've got the best partner in the [39:18] world on this, you know, Michael Dell. [39:20] >> Yeah. [39:21] >> Yeah, joined me he and Susan joined me [39:23] on on this journey. Really helped me [39:26] over to get it over the last one inch [39:28] line [39:29] >> Yeah. [39:29] >> with the administration and then made [39:31] the biggest philanthropic gift in [39:32] history, 6.25 billion. Um, $250 to 25 [39:37] million kids. [39:38] >> Yeah. [39:38] >> And you know, frankly I think for [39:40] Michael and Susan they're just getting [39:41] started and I think their example that [39:44] they've set for everybody else. Um, you [39:47] know, if you have if you you know, you [39:49] guys look at the amount of wealth that's [39:50] being created here in Silicon Valley. I [39:52] mean it it's it's it's it's really there [39:54] is no historical precedent. There is no [39:56] historical precedent. Um, and the fact [39:59] of the matter our charitable our [40:00] charities are not prepared or equipped [40:03] to take 10, 20, 50 billion dollars. Like [40:07] and a lot of people want to give away [40:08] this money during their lifetime or you [40:10] know, within 10 years of dying. [40:12] >> and and targeted too. [40:14] >> Right. [40:15] >> And in a way in a way that there's no [40:16] >> And scales up or down. It's like you can [40:18] do the whole state. You can do your [40:21] county, you could do school. [40:24] And there's not 30% overhead on the [40:26] charity where somebody's getting paid [40:27] $10 million and you know, all this stuff [40:30] happens after you pass away. 100% of it [40:32] goes directly to the kid. [40:34] >> Yeah, charity was so vague for so long. [40:35] It was like, great. Okay, you gave away [40:37] half your money, but you're actually not [40:39] transferring until you die and then it's [40:41] going to go into this charity that will [40:42] deploy it later. It It gets so abstract [40:45] that I think people [40:46] all of those big donations that happened [40:48] in like the previous era sort of fell on [40:50] deaf ears and they weren't didn't feel [40:52] like they were moving the needle. And so [40:54] this is just an entirely new way to do [40:55] it. I love it. [40:56] >> The chat is asking if you have any surf [40:58] trips planned. [40:59] >> Wow, wonder who that must be checking [41:02] out my [41:03] my Twitter picture. [41:04] Which by the way was at the surf ranch. [41:07] With Ramondo. And the picture actually [41:10] some people think it's me, it's not me. [41:12] That was my then 11-year-old son. [41:15] >> No way. [41:15] >> Getting barreled at the surf ranch [41:17] because Ramondo was like telling him how [41:19] to get into the barrel. [41:22] But I have to say I'm 55 guys. I just [41:25] had a birthday. I'm working harder than [41:28] >> I feel [41:30] like I just can't imagine you being [41:32] like, yeah, now's a good time to take a [41:34] surf trip. I feel like maybe a trip to [41:37] surf ranch, but [41:38] >> Yeah, you got to [41:40] >> We we ought to get to we ought to get [41:41] together and do that. By the way, I'm [41:43] I'm I'm currently signing up somebody [41:45] who's going to [41:47] adopt all the kids in Los Angeles. We've [41:48] got San Francisco already covered. We've [41:50] got Oakland already covered. [41:54] We're going to announce some big things [41:55] here in the state of California. I'm not [41:56] giving up on California. [41:58] >> Yeah. [41:58] >> Right? We're going to defeat We're going [41:59] to defeat the unconstitutional taking [42:02] tax. [42:02] >> Yeah. [42:02] >> That some people call the wealth tax or [42:04] the billionaire tax. Like this attack on [42:06] success, [42:08] you know, trying to divide wedge and [42:10] wedges between Americans. We're uniting [42:12] people with the Trump accounts, with the [42:14] Invest America accounts. We're raising [42:16] the floor and getting everybody into the [42:17] game. And this whole idea that we're [42:19] going to demonize success and drive Elon [42:21] out of the state, etc. Uh shout out by [42:24] the way to my junior [42:26] uh son, Lincoln Gerstner, [42:28] who published his first paper this week [42:31] and I I show up at home and it's it's on [42:33] the economic impact of tax policy in in [42:35] California. I show up at home and he [42:37] said, "Hey Dad, I I I I I finally uh you [42:40] know, posted that that paper I was [42:43] writing." He's doing it with Joshua, the [42:45] incredible professor over at Stanford. [42:47] He's And then he says to me, he's like, [42:48] "Has Marc Andreessen ever retweeted [42:50] you?" [42:53] I said, "No. I was like, [42:55] I don't I I don't think so." And he [42:56] goes, "I think he retweeted me." [42:59] And I was like, "No, he definitely [43:00] didn't retweet you, but Marc did. So, [43:02] shout out to Marc." [43:03] >> That's awesome. [43:04] That's awesome. [43:05] >> And you know, I I I think that we are [43:07] what what people there's a lot of [43:08] despondency in California. [43:09] >> Yeah. [43:10] >> Um [43:11] I'm going to take a contrarian position [43:13] here. [43:14] Spencer Pratt's going to be the new [43:16] mayor of Los Angeles. [43:17] The wealth tax will be defeated. [43:20] We we will pass the retirement and [43:22] personal asset protection act as a [43:25] referendum in California, which will [43:26] prohibit people from stealing your [43:28] retirement money or your personal [43:30] assets. That will get passed, okay? [43:33] That will send a shocking message to the [43:36] rest of America. The rest of America [43:38] thinks that California is as blue as it [43:41] gets. It turns out California's pretty [43:43] purple, [43:44] right? And I think that common sense [43:46] initiatives are going to [43:49] uh you know, reassert themselves in uh [43:51] you know, in the election in November. [43:53] Um and I think it's great because we're [43:55] the fourth largest economy in the world. [43:56] I know some of my friends moved out and [43:58] said, "Listen, California's got it [43:59] coming to them." My own view is this, as [44:02] California goes, so goes the country. We [44:04] cannot cede California. It is where [44:07] we're going to battle for the best ideas [44:08] that are consistent with the founding of [44:10] the country and we're going to, you [44:12] know, win on those ideas. And so I think [44:14] we're we're seeing a lot of progress. [44:15] Shout out to Sergey [44:17] and building better California and the [44:18] incredible work that they did to to get [44:20] us moving in the right direction. [44:21] >> It's fantastic. [44:23] >> Well [44:23] >> kept you away too long but thank you so [44:25] much for coming to the show and hanging [44:26] out. [44:27] >> Uh and excited for your next project. [44:29] >> Yeah. [44:29] >> How How much did you sell How much did [44:31] you sell this for? [44:33] >> Let's go for a Let's go for a Let's go [44:35] for a serve. [44:36] >> Yeah, we can only save far away from the [44:37] mic. [44:38] >> I'm going to I'm going to turn this into [44:39] a little BG2 and turn the tables on you [44:41] guys. I need to get some more some more [44:43] the other way. [44:44] >> Great to see you. Have a great weekend. [44:45] >> see you, Brad. [44:45] >> Thank you. [44:46] >> You're the man. [44:46] >> We'll see you again. [44:47] >> Cheers.