第 298 期·中文精读

Anthropic 上市,英伟达陷入危机,市场暴跌,开源监管收紧?

Anthropic Goes Public, Nvidia in Crisis, Markets Tank, Open Source Crackdown?

Anthropic 上市引发多空激辩,英伟达面临股价、人才与法律危机,白宫拟收紧中国开源模型披露,查马斯提出十大AI巨头论。

官方音频00:00 / --:--

本期章节点时间可跳到音频对应位置

  1. 0:00查马斯:10 家 AI 巨头的“从底层到应用”理论
  2. 3:27Anthropic 上市,企业如何应对 AI 颠覆
  3. 16:41Anthropic 与市场的多空逻辑
  4. 36:41白宫考虑对中国模型出台新的开源披露规则
  5. 59:17英伟达现状:股价暴跌、人才危机、法律问题与黄仁勋的未来
中文精读由 AI 基于原始节目改写,非官方翻译,可能存在偏差或错误;观点归属与表述以原音频与官方文稿为准。说明

英伟达刚公布的季度财报里,营收 1160 亿美元,净利润 650 亿美元,两个数字都在已经极其庞大的基数上翻了一倍。美国股市上大多数公司的总市值,还比不上英伟达这一个季度赚到的利润。然后股价跌了。三十天里市值蒸发 2 万亿美元,跌幅 35%。

同一天,Anthropic 上市,估值 2 万亿美元,融资 1070 亿美元,绿鞋全额行使。这是人类历史上最大的一次 IPO,超过了 SpaceX 此前的 850 亿美元。市场没有因此普涨,反而是一片血红:标普 500 跌 1.39%,VIX 涨 10.83%,AMD 因为一份做空报告盘中跌了 10% 到 12%。

这一期 All-In,萨克斯缺席,杰森·卡拉卡尼斯(Jason Calacanis)、查马斯·帕里哈皮蒂亚(Chamath Palihapitiya)、大卫·弗里德伯格(David Friedberg)三个人从 Anthropic 上市聊到英伟达危机,中间穿插了一条关于开源监管的线索。查马斯开场就抛出一个理论,后面几乎所有讨论都挂在这个框架上。

未来十年只会剩下十个巨无霸,每一个都从土地一路吃到应用

查马斯的判断是:AI 最终会催生九到十个巨型公司,每一个都做全栈垂直整合,从最底层的土地、电力、数据中心,一直做到最上层的应用。

他的证据是前一天 Palo Alto Networks 的收购。这家公司本该老老实实待在安全赛道,却买下了一支做开源模型的团队,股价当天一度涨了 8% 到 9%。CEO 尼凯什·阿罗拉(Nikesh Arora)在新闻稿里直说,这是他们第一次大规模走出安全领域。市场的反应是:谢天谢地。

他先拥有模型,然后往下拥有土地和数据中心,再往上做应用层。Anthropic 拿着那么多钱在干什么?已经在往土地走,也在往应用走。OpenAI 一样,SpaceX 一样。他们手里有 Cursor 这样的顶级应用,也一直挖到最底层。(查马斯·帕里哈皮蒂亚)

Meta 也是这个结构。查马斯说,AI 正在制造有产者和无产者的分野,有产者就是那些握有关键资产、能够 tightly 垂直整合的公司。剩下的生态里会有大量开源模型、一些闭源模型,以及大量分布式算力。

这套逻辑顺带解释了英伟达为什么在跌:它超额盈利的时代可能结束了。算力的形式和路径变得多样,真正稀缺的约束变成了土地和电力。市场眼下赌的是那十个巨无霸席位的归属,芯片性能只是其中一项筹码。

那些手里握着大把自由现金流、却不打算进模型或数据中心游戏的老牌公司,会遭遇什么?查马斯的说法是,市场不会直接说你的股票不好,它会用脚投票,把钱挪到那些愿意承担风险的公司里去。

回购是"我不知道该买什么"的公开自白

杰森提到苹果过去十年的选择:股本减少了 34%,Alphabet 减少 12%,Meta 减少 11%。蒂姆·库克治下的苹果,很大程度是靠大规模回购和分红撑起来的。

查马斯对这个动作的评价很不客气。

当你宣布回购,你实际上是在说:我也不知道该买什么,不知道该建什么,不知道该投什么。扰动低的时候这没问题。扰动高的时候,市场会奖励并购,因为并购在告诉现有股东,你在努力守住护城河、建新的护城河。(查马斯·帕里哈皮蒂亚)

他算了一笔账:过去很多年美股 EPS 的增长,很大一部分来自这个游戏。利率低,公司有利润,就把自由现金流拿去买自己的股票,对冲掉发给员工的股权激励造成的稀释,每股收益自然上去。股东心里有数,公司每年买掉一部分流通股,股价大概率每年涨 5% 到 10%,这个游戏可以一直玩下去。

现在这个游戏结束了。因为它是 a tell,一个暴露底牌的动作。

杰森提了一个反驳:回购在股价被严重低估的时候是合理的,或者你该比较一下,买自己股票和买另一家公司哪个回报更高。查马斯抓住了他用的一个词:accretive,增厚。他说这个词是电子表格玩家的思维,只算今天这笔交易对每股收益有没有增厚,不算这笔交易能不能在几年后把你带进今天不在的业务线里。

如果你被一个不给你这种许可的股市绑住,你的处境非常麻烦。它不奖励你回购,也不奖励你分红。你手里这笔钱能干什么?只能投出去,投出一条路来。(查马斯·帕里哈皮蒂亚)

杰森补充了一个现实约束:并购能不能做,跟华盛顿的态度直接相关。拜登时期大并购基本被挡住,特朗普时期重新打开,但如果 2028 年民主党上台,这个窗口可能只有六个月。所以要在场上打球,规则是随时会变的。

Anthropic 上市:2 万亿美元,30 倍市销率

Anthropic 的 IPO 数据:估值 2 万亿美元,融资超过 1070 亿美元,绿鞋全部填满,据说再来一百个绿鞋也能填满。营收在 800 亿到 1000 亿美元的量级,市销率大约 30 倍。

杰森说这个时点选得聪明,在市场动荡中把门撞开,先出去再说。他关心的是另一件事:AI 交易里最大的一块,也就是 token 销售和前沿模型,此前一直藏在私募市场里,现在终于被拖到阳光下,可以和 SpaceX、AWS、Azure、英伟达、Cerebras 放在一张图上看清楚了。

弗里德伯格从内部听到的消息是,增长超出了他们自己的预期,1000 亿美元营收的目标看起来轻松可达。企业级产品是市面上最好的,比 Google 和 OpenAI 都强。

他提了一个概念:RSI,递归自我改进。Google 最近几周连发了四个模型,很多人私下议论它已经进入了这个循环。如果 Google 到了,Anthropic 大概率也到了,只是时间问题。衡量方式是"每美元能买到多少智能",如果这个比率在持续指数级改善,那对这类公司来说,天花板就是整个经济。

他们在路演里做得很漂亮,说服了投资人相信这门生意有一种近乎天然的垄断优势:递归自我改进加上单位智能成本的持续下降,构成一道别人追不上的护城河。最大的风险是开源。(大卫·弗里德伯格)

但他认为开源吃不掉整块业务。如果 Anthropic 在高端应用上始终提供最优的"每美元智能",它就永远是那个选择。真正的赛道是 Anthropic、OpenAI、Google 三家。

他更关心的是流通盘结构:今天买进的 1000 亿美元里,有多少是短期持有者,有多少是真打算长拿的。他估了个数,大概一半。也就是说真实流通盘约 500 亿美元,对应着数倍于它的需求。如果愿意拿五年以上,从今天的入场点看,这可能是一只多倍股。同样的判断也适用于 SpaceX,甚至当下的 Google。

查马斯泼的冷水:未来三四个月,Anthropic 的股价和 Anthropic 关系不大

查马斯同意 Anthropic 会成为那十个巨无霸之一,做的事都对。他唯一担心的资产负债表问题,这次 IPO 之后已经 de-risk 了。

但他对估值本身不表态,理由很直接:一家公司再好,也扛不住宏观。

看英伟达,我们从来没见过那样的一个季度,股价却在跌,而且从财报日算起跌得不少。为什么?市场看起来是在硬撑着往上走。问题是通胀很黏,所有外部条件都没解决,包括战争。大家心里发虚,觉得这轮行情跑了很久了。天平是往降风险那边偏的,不是往加风险那边偏的。(查马斯·帕里哈皮蒂亚)

他列了一串压在股票上的东西:十年期国债收益率还在往上走,波动率高企,通胀黏性,凯文·沃什(Kevin Warsh)就算想降息也没多少空间,搞不好还要加息,参众两院换了新主人带来不确定性。AMD 因为一份说它向中国卖芯片的做空报告,一天跌 10% 到 12%,这可是万亿市值的公司,这种波动幅度是巨大的。

所以他现在的姿态是 risk off。他看空的不是 AI 本身,他担心接下来还有更多东西要出来,分三块:经济层面、AI 交易的一阶效应(要等国会听证会开完才知道)、以及整体环境给人的感觉。最后一条他讲得很具体:每天都有网络攻击,很多东西停摆,官方让大家准备三天的应急物资,这种氛围下,风险的天平难道不是朝下吗?

杰森不服,说这就是在底部附近震荡,除了黑天鹅之外没什么可怕的。两个人就"市场是不是见底了"来回拉锯了好几轮,查马斯的核心反驳始终是同一个:你没法从"token 用量会涨 500 倍"推演出"股市见底了",十年期国债在 5.5% 还是 4.5%,对全部股票的估值有数学上的直接含义。

杰森的 500 倍:初创公司已经在这么用了,大公司两三年后会跟上

杰森的看多逻辑建立在他自己的观察上。他一年投 100 家初创公司,看到的是这些公司的员工每天和 AI 协作五到七个小时,比和人类同事协作的时间还长。而大公司里只有 10% 到 20% 的员工在用。

初创公司今天做的事,大公司两三年后一定会做。当年拥抱云计算是这样,用 WeWork 是这样,因为效率更高。所以我看到的是使用人数会翻五到七倍,更重要的是使用时长会从每天几分钟、一小时,变成六到八小时。你会和 agent 一起工作,你会同时协调十个虚拟员工。这是显而易见会发生的事。(杰森·卡拉卡尼斯)

他的推论是 token 用量至少涨 500 倍。这些 token 从哪来?Anthropic、SpaceX、开源、AWS,全都有份。

他还给了个更硬的算法:知识工作的人均营收会从 50 万美元走到 300 万到 500 万,甚至 1000 万美元。微软的人均营收从 90 万走到 200 万,这个趋势每三年翻一次。如果人均产出翻倍,公司根本不在乎员工在软件上多花十倍的钱,因为软件本来只占人均成本的 5000 美元,涨到 2.5 万也无所谓,换来的是一个超人。

查马斯不否认使用量被低估了,他只是不认为这能推出股价见底。他说市场现在是"show me"状态,要看到真东西才肯付钱。

弗里德伯格插进来,把讨论拉到了另一个维度。

我认为 AI 最大的机会在前沿,也就是企业和个人能做到人类和商业此前从未做到的事:解决真正困难的工程问题、科学问题、以前啃不动的难题。那才是 100 倍、1000 倍的空间。而不是拿 AI 去迭代现有业务流程、省钱、提效。大多数人不在前沿工作,所以他们的第一反应永远是"这会怎么改变我的工作",然后立刻开始恐惧。(大卫·弗里德伯格)

他的观点是,把 token 扔向现有流程,回报率可能只有 50%、60%;把 token 扔向没人做过的事,回报率是另一个量级。顺带他也把"工作末日论"和"人人要靠 UBI"的说法判了死刑:历史上每一次前沿时刻,新工作、新产业都在旧工作消失之前就被创造出来了,所以失业率没暴涨,所以大家没排队领 UBI。旧金山当年成为美国人口第二多的城市,是因为所有人都在淘金热里往前沿跑。

但他也承认有个 K 型问题:不是所有人都上了这趟车。没上车的人看着前沿在往前跑,自己不在里面,反弹就来了。这种反弹有可能反过来掐死前沿本身。

白宫在讨论给用中国开源模型的公司"贴标签"

杰森把话题转到监管。路透社报道,白宫正在讨论一项新的开源披露规则:公司可能被要求在官网上披露自己是否使用了中国的开源 AI 模型。推手据说是财政部的贝森特(Bessent),他正在成为对华鹰派的领头人,另一边是 OSTP 的迈克尔·克拉齐奥斯(Michael Kratsios)。白宫内部正在打一场内战。

今年早些时候,Airbnb、DoorDash、Cursor 都收到了众议院中国特别委员会的信,理由是它们使用了中国开源模型。Airbnb 的回应是:对,我们的客服用了。

杰森自嘲说,我还在用 MacBook Pro 做这档播客,是不是也得贴个标签说我们用中国硬件生产内容。

查马斯的回答很短。

在网站上贴个免责声明说你在用开源模型,会有什么后果?没有任何后果。它只会说明你不傻。(查马斯·帕里哈皮蒂亚)

他接着把逻辑推下去:市场想要什么?想要你聪明、省钱、讲 ROI、技术自主。它不想要你把 alpha 泄露出去。反过来,如果你说"我不在乎成本,什么模型都行,我把所有 trace 和 span 都喂给它,它也许会学着怎么跟我竞争,但他说零数据留存,我信他",市场会觉得你是个白痴。而且你会证明自己确实是。你会暴露你团队里连半个懂技术的人都没有。

所以这是个二选一:贴一个没人看的免责声明,还是关门。当然贴。

他真正的结论是,华盛顿需要真正懂技术的人在内部,把这些东西的技术含义讲清楚。这个提案是个试探气球。如果真的一条条推演下去,会发现不让企业用更便宜的开源方案保持竞争力,对美国公司的伤害更大。难道我们想要的是相反的声明,声明自己完全不用模型?

他顺手把开源这件事的常识讲了一遍:你并没有从中国人手里买任何东西。它在 GitHub 上,在一个网站上,你下载下来,fork 一份,放到自己的服务器上,你控制它。没有后门。就算有,立刻会有一千个 AI 盯着那个后门。这跟用华为的硬件完全是两码事。

弗里德伯格:给软件贴标签,是通往暴政的路

弗里德伯格这一段讲得最重,值得完整引。

任何试图遏制开源、限制开源、给开源贴标签、给政府提供工具去制造事实上的受监管垄断的努力,都会限制人类的繁荣,限制个人的自由,制造一个奥威尔式的国家,而且从根本上侵犯言论自由。我不认为开源软件的来源国应该成为一个问题。(大卫·弗里德伯格)

他给了一条历史线索。Apache 基金会做出了开源服务器,由此有了开放的互联网;Mozilla 和 Firefox 让网景和微软没法限制人们上网;Google 做了 Android,找到硬件厂商把它装成事实标准,今天全世界 70% 的手机因此能不受限制地接入互联网。在那之前,闭源软件公司想控制你能装什么应用、能访问什么网站,一切都被监控,一切都被控制。

现在开源 AI 面临同一个决策点。

每一个掌握了"什么可以看、什么可以做、什么可以访问、哪些权重算事实、哪些算已被证实的科学、哪些算真相"的权力集团,最终都会变成暴君。他们不是有意要当暴君。但他们对这些事有自己的看法,认为自己是对的,一旦手里有权力,就会照着这个看法去行动。(大卫·弗里德伯格)

他还纠正了一个技术事实,这个纠正很关键:开源权重不是一个可执行程序。你下载下来,它不会在你电脑上跑起来,也不是什么能黑进你机器、偷走你数据的隐藏工具。它是一堆数据,一个装满了数字的 CSV 文件。你需要一个运行环境才能拿它跑推理。所以"我们不知道中国人在里面藏了什么"这个说法,在技术上是不成立的。所有东西都摊在明面上,你可以审计、可以测试、可以做安全筛查。而且你必须做,因为一旦你基于它做出来的产品出了问题,责任是你的。

他还指出了"标注来源国"这件事的荒谬:在 Hugging Face 上,任何人都可以匿名发一个模型,它可能是史上最好的开源模型,上面不写它来自哪个国家。它是在某个地方训练的,某群工程师写的,发出来给全世界用。你怎么分类?来源国为什么重要?

查马斯追问了一句:那些现金充裕的上市公司,现在开始在公司内部搞开源模型项目,你怎么看?弗里德伯格的回答是百分之百支持,他认为这就是 Meta 该走的路。杰森补充说 Google 也在做,但都没做到前沿水平,Meta 的开源模型被 nerf 很久了。查马斯接了一句:已经被 nerf 很久了。

为什么能打的开源模型全是中国的

这是查马斯在这一段里问出的最有价值的问题。

回头看,本来有一些事是可以做的,能给开源做出真正的逃逸速度。美国公司手里有那么多钱,为什么没找到办法花一部分出来,做出一个真正的美国开源替代品?为什么唯一能打的开源模型都是中国的?为什么?(查马斯·帕里哈皮蒂亚)

杰森的答案是人才流向。人才面前有几个选项:去做闭源模型,个人能赚到几亿甚至几十亿美元;或者去 OpenAI、去 xAI。钱都往那边流,人也是。

查马斯不接受这个解释,因为 Meta、Google、英伟达里都有一流的人。他真正的问题是:在一万亿美元现金里,为什么没花出 500 亿或 1000 亿,做出一个像样的替代品。

弗里德伯格的答案是资本市场结构。过去两年,以"我们要建一个前沿实验室"为名融资,比以"我们要建一个开源 AI 公司"为名融资容易得多。你去募资,讲开源,投资人第一句问的是怎么赚钱,你说以后再说、为了人类福祉,这个对话就结束了;讲闭源,标准商业模式清晰,钱就来了。所以是资本家整体塑造了这套激励。

而中国的博弈结构完全不同:他们想的是怎么打乱美国的 AI 生态,怎么造出一个不由美国控制的替代品。对他们来说,做开源权重模型是非常理性的选择。

查马斯追问:那同样的博弈逻辑为什么不适用于英伟达、Meta、亚马逊?杰森接得很快:英伟达确实在这么干,买了 Hugging Face 和 Poolside,花了 200 亿美元,把开源资源全收了。落后的一方去拥抱开源,去杀死已有的封闭方案,这就是当年 Android 对 iOS 干的事。

查马斯说,那些收购都是关于托管模型的,不是模型本身。真要下注,多给 Nemotron 一些算力就行。他看了 Nemotron 的 checkpoint,落后了。原因可能是算力受限,也可能是人才受限。模型不差,但不够好。GLM 和其他中国变体比"好"更好一点,也还没到"很好"。真正"很好"的是那几家前沿的。

杰森提到一个细节:现在美国这边能期待的开源力量是 Muse Spark、Nemotron,还有 Reflections 的 Beam,都排在 Kimi、DeepSeek、GLM 后面。中间还夹着一个法国的 Mistral,位置刚好在美国那三个落后选手和中国那三个领先选手之间。他预测一年后,美国这几家会落在 DeepSeek 和 Kimi 中间。

查马斯最后给了另一个可能的路径:现在有可能发生,因为前沿的人正在分裂。前阵子有一篇叫《Misanthropic》的东西,一堆签名者,都是有效利他主义圈子里的人,出现了裂痕。完全可能有一批在模型公司赚够了钱的聪明人离开,去做非营利的开源替代品,因为他们有钱,本来也不太资本主义。杰森补充说,《纽约时报》刚做了个达斯汀·莫斯科维茨(Dustin Moskovitz)的大报道,讲他在 Anthropic 和 OpenAI 的早期投资,以及他对这一切的不安。这些人的钱和不安加在一起,做开源替代品的概率不是零。

英伟达:财报炸裂,股价崩盘,创始人出局

杰森把最后一段留给了英伟达,数据摆得很清楚:三十天蒸发 2 万亿美元市值,跌 35%;詹森被"捅刀"之后两周跌 20%。CFO 兼代理 CEO 科莱特·克雷斯(Colette Kress)的第一次财报电话会没给人多少信心,但财报本身是炸裂的:营收 1160 亿,净利润 650 亿,双双击穿预期,都在巨大基数上翻倍。

股价照跌。

然后是几个具体动作。克雷斯说公司会在新云(neo cloud)融资上"更有纪律、更有选择性",杰森的说法是"詹森银行关门了"。H200 对华销售也停了,虽然商务部一月才批准。鲁宾和布莱克威尔本来就禁售。董事会反复表态詹森不会回来,詹森自己说愿意谦卑地回来,外面有一群人在推他复出。

法律层面:SEC 调查、集体诉讼,都指向可能误导了投资人的陈述;DOJ 在继续查与 Megaspeed 的关系。人才层面:员工签请愿书,特拉维斯·卡兰尼克(Travis Kalanick)公开建议罢工。他讲自己当年被赶下台时撑了六个月,"请愿不错,罢工更好",还说愿意带着一个镶满晶圆的扩音器去英伟达总部,跟想罢工的员工和相信这件事的创始人站在一起。"结果我们控制不了,但先打一仗。"

查马斯对这件事的态度,是这一期里最克制的一段。

他是一位非凡的、传奇的创始人和 CEO。他建起了世界上最重要行业里最重要的公司。他好到不真实。他也犯了一些错误,错到什么程度现在还不知道。问题在于,你不能把这样一个人重新塞回一家市值被摧毁成这样的上市公司,那等于把公司治理整个扔出窗外,让所有人承担风险,包括公司本身。所以这事不会发生,所以董事会必须这么表态。(查马斯·帕里哈皮蒂亚)

他觉得克雷斯很优秀,只是被放在了一个尴尬的位置上,充当传话人,然后被网上的喷子围攻,这不该由她承担。

他反复强调的核心是时间点:12 月 1 日和 2 日,这三拨人要面对三个不同的委员会,特别委员会、众议院情报委员会,还有一个。这些场合一点都不轻松。

所有人先冷静下来。他会非常专注地准备 12 月 1 日。希望结果是"这里没什么可看的"。但我可以保证,只要有一丝一毫"这里确实有东西",它就会泄漏出来,然后又是一波,因为赌注太高了。互联网暴民、对华鹰派,谁知道还有多少方势力在场。这就是为什么我说这件事在 too hard 的筐里。大家越是就这些表面的东西瞎嚷嚷,他回来的难度就越大。(查马斯·帕里哈皮蒂亚)

弗里德伯格的立场更极端:他不相信任何报道。

发生了什么就是发生了什么,他辞职这件事改不回来了。我们不知道他辞职是因为对华售芯的顾虑、是因为间谍问题、还是因为私德和违反行为准则。公司没有给出定论,詹森没有给出事实陈述,政府没有给出任何事实声明,没有任何民事或刑事起诉。我们全都只是在对着媒体反应,而我们平时一直说媒体在骗我们。所以我不知道该信什么。(大卫·弗里德伯格)

杰森试图引用报道,被弗里德伯格一次次挡回去:"那是媒体报道""某个记者说的""我不再对媒体做出反应了"。

杰森最后还是给了自己的猜测:看所有报道的整体,再结合他对销售这个群体的了解,他赌最后会查出来是一群灰色地带的销售、层层转手的经销商,把货倒给了中国公司。查马斯回了一句:你应该被传唤去 12 月 1 日做特别证人。

这一期真正的分歧在哪

三个人在事实层面几乎没有分歧:AI 的使用量被严重低估,Anthropic 会成为巨头,英伟达的财报好得离谱,开源不该被贴标签。

分歧在时间尺度上。

杰森看的是三到五年,他看到初创公司每天用 AI 五到七小时,看到人均营收翻倍的趋势,结论是现在就该买。查马斯看的是三到四个月,他看到的是十年期国债、波动率、通胀、政治换届、国会听证会,结论是 risk off,好东西会更便宜。

弗里德伯格看的是十年以上,他关心的是人类能不能用这项技术去解决以前解决不了的问题,以及会不会有一场无声的垄断把这条路堵死。

有意思的是查马斯在结尾用市场数据支持了弗里德伯格的立场。他念了当时的盘面:标普跌 1.39%,VIX 涨 10.83%,Anthropic 较发行价涨 18%,而全场唯一在涨的另一家公司是 Palo Alto Networks,涨 5.86%,因为它押了开源。

弗里德伯格是对的,市场会奖励你。市场在说的是,这在哲学上是对的,在数学上也是对的。在这样一天里,这个信号分量很重。(查马斯·帕里哈皮蒂亚)

所以他同时持有两个判断:整体风险偏下行,但奖励那些敢下重注、敢做垂直整合、敢押开源的公司。这两件事不矛盾,它们描述的是同一个分化的市场。

  • 看财报要看现金流,别看标签。软件公司还是服务公司这个分类正在失效。用毛利率和自由现金流转化率筛公司,比用行业标签筛靠谱得多。一家 99% 毛利、70% 自由现金流利润率的服务公司,比任何产品公司都值得持有。
  • 把大规模回购当成一个负面信号来读。回购本身谈不上错,问题出在时机:当一家公司在高扰动时期选择回购,放弃并购和新业务,它等于公开承认自己找不到值得下注的方向。这种公司在新周期里会被持续抛售。
  • 如果你在创业或做投资,现在该问的问题是"AI 让我能做哪些以前根本做不了的事"。省成本那一侧的 ROI 只有 50% 到 60%,前沿那一侧是另一个量级。这也是为什么失业率没暴涨:新东西的吸引力远大于重造旧东西。
  • 关注 12 月 1 日和 2 日。英伟达那三个国会委员会听证会,是这条线上所有不确定性的收敛点。在那之前,任何关于詹森复出、关于调查结论的判断都是噪音。查马斯说得直白:那时候才知道"詹森银行"会关到什么程度。
  • 开源监管这件事,值得当成一个长期变量盯着。要求披露是否使用中国开源模型,听起来只是个声明,但它可能是给闭源产品制造受监管垄断的第一步。这条线往后走,会影响你选技术栈的成本结构,也会影响你公司的估值逻辑。
  • 对个人来说,最实际的一条:现在就把 AI 用满六到八小时。杰森看到的差距不在模型能力,落在使用强度上。初创公司员工每天和 AI 协作五到七小时,大公司只有 10% 到 20% 的人在用。这个差距会在两三年内变成人和人之间、公司和公司之间的差距。
文稿来自 All-In 官网(allin.com)发布的官方 Transcript,说话人按官网分轨数据标注,个别段落可能张冠李戴。
Jason Calacanis (00:00:00): Alright, everybody. Welcome back to the number one podcast in the world. Seven weeks running. That's right. We manifested it. We got a crazy docket. Sax couldn't make it this week. He's out. Shammoth, you were a little on fire in the group chat. You have some theories. You have some thoughts. Please, the floor is yours. Chamath Palihapitiya (00:00:15): I have a theory. Jason Calacanis (00:00:16): You have a theory. Okay. Just one, but it's a big one. Let's let's let's unpack it. Chamath Palihapitiya (00:00:20): I think that there are going to be nine to 10 mega corps that get built over the next decade because of AI. Speaker 3 (00:00:27): Yes. Chamath Palihapitiya (00:00:27): And I think that each mega corp is gonna go from the application layer all the way down to the dirt. Now, do I say this? I think the deal that Palo Alto Networks did yesterday rips the Band Aid off, and I think was very smart. What did they buy? They bought these brilliant guys that made open source models. And you would say, well, why is a security company, which should narrowly only be allowed to be in the security lane, why were they rewarded yesterday, the stock was up eight or 9% at one point, for buying an AI company. And Nikesh said it in the press release, he's like, this is us venturing outside of security, our first big venture. And the market was like, thank God. And so what's happening? He's gonna own the model, right? Then he's gonna eventually own the dirt and the land and have the data centers vertically integrated. Then he's gonna go up the stack applications. What does Anthropic do at Ant with all their money, they're already going down to the dirt and they're going up to the application stack. OpenAI doing the same thing. SpaceX doing the same thing. They own one of the best apps in the game, Cursor. They own all the way down to the dirt. So what do you see taking shape? Meta. Meta owns all the way down to the dirt, all the way up to the app stack. So I think what AI is creating essentially is haves and have nots. The haves are the folks that have these critical assets that can be very tightly vertically integrated. And I think that there's gonna be 10 of these folks. And so you're gonna have this really rich app ecosystem, a lot of open source models and some closed source models, and then you're gonna use a lot of distributed silicon. This, I think, is why NVIDIA is trading down, is this realization that the over earning era of NVIDIA may be over. This is independent of Jensen, where you just have a multitude of formats and ways in which you can actually compute what you need, and then you're gonna need this really important constraint, which is the land and the power. So I think if you view it that way, Jason, you're gonna see these eight, nine, 10 conglomerates and now the market is betting who is a conglomerate and who is not. Speaker 3 (00:02:32): And Chamath Palihapitiya (00:02:33): that more than anything I think explains the flows over the last few weeks and probably like the last year even. Interesting. So who's on the back heel that has to figure it out? If you are an established company with large amounts of free cash flow and you're not gonna go and enter the model game or the data center game, I think the market's going to punish you not by saying your stock is not good, but essentially by trading out of your stock to trade into those companies that are willing to take the risk. Jason Calacanis (00:03:04): You need something that's more ambitious than Microsoft clearly is gonna need to have a frontier model. Chamath Palihapitiya (00:03:11): It doesn't mean anything. What does it mean to make a product in a world where you have a trillion agents that you can spin up on demand to make anything you want? What is a product? I don't know what a product is. What is that? Jason Calacanis (00:03:21): It is a valid question if you can add new features at the rate Ant is doing. And speaking of mega corporations, Ant went public today. Anthropic, Freebird just upsized their IPO. Massive demand. They hit a $2,000,000,000,000 valuation. Raised over a 107,000,000,000 green shoe filled. A 100 other green shoes could have been filled apparently. That surpassed SpaceX's $85,000,000,000 IPO. And every other corporation seems to be off a bit today as people rotate into this. I guess retail really wanted to get a piece of the anthropic IPO with all the hype and people understand the product, I think. And this is gonna be, I think, Freeburg, you might agree. We've been waiting to have some sunlight on a big portion of the AI trade, of the AI revolution, which was tokens and token sales and frontier models. And now we're gonna have all of that exposed to the light, and we're gonna be able to figure out between SpaceX and other cloud providers, Amazon Web Services, Azure. Now we have this missing piece. We're gonna kinda under... And we... Obviously, we have NVIDIA to discuss as well today. We'll have NVIDIA's, you know, chip sales, Cerebras' chip sales. Now we understand, like, the the complete picture, but are you... What's your what's your take on this... The largest IPO of all time? And then maybe a little bit on Chamotz's New Theory, eight mega corps, 10 mega corps, and you gotta be in everything if you can't build software anymore because AI will build it instantly. There's no value or mode around it. What's your take on this mega corp dirt to services, I Speaker 3 (00:05:07): guess, theory. Chamath Palihapitiya (00:05:10): Dirt to application. Jason Calacanis (00:05:12): Yeah. Not you wouldn't include services in there? Because it seems like services in these forward deployed engineers... Chamath Palihapitiya (00:05:17): You're bringing up something that's really important. This whole idea of software and services is for Luddites of the past. And all these people, look, I think the problem when things change is some people find so much anxiety and change that what they need to do is they need to glom onto the past, they feel better about the past and they use that to self soothe themselves. Jason Calacanis (00:05:45): It's a pacifier, yeah. Chamath Palihapitiya (00:05:47): What is a version of that in the public markets? I think it's this thing that they used to do to bucketize companies. And again, I will bring up that very famous memo about what you report when you can't report free cash flow. Does anybody actually care whether you're a software company or a services company? You shouldn't. The people that care are morons. What you should care about is what is your free cash flow. Like if I could sell you a services business that had 99% gross margins and 70% free cash flow margins, would you buy it? Your answer would be yes, you'd buy it. And you'd buy it over every product company that ever existed because it's just structurally, financially better. And in a world where AI is gonna be disrupting everything, what do you wanna own? I would rather just own cash flow. And so I think what's gonna happen, Jason, is people are gonna stop caring about what kind of business you are from a business model perspective and care a lot more about what kind of business you are from a numerical perspective. And if you have high gross margins and high free cash flow conversion, you will be rewarded. And if you don't, it won't matter what the hell you do, you're just gonna get decimated. Jason Calacanis (00:06:53): So the theory is the markets are gonna be confused here. AI is shaking up the snow globe. Existing businesses and verticals, hard to... For maybe people who are living in the past paradigm, they're trying to say this is a services business, This is a software business. This is an infrastructure business. Now, it's all one giant thing. And SpaceX kinda drove this by putting all these pieces together and when they put Tesla and SpaceX together, now you've got a conglomerate. Google, obviously, in conglomerate territory with Waymo on one side and their cloud business, their advertising business. And so... And Amazon as well, I guess, has a conglomerate type feel to it. Microsoft maybe less so, and Apple maybe less so. Right? Because they kinda stay in their lanes. Apple consumer hardware, Microsoft business software. You know, they got a little Xbox business over here that they've threatened to spin out. So what do you think of this theory, Friedberg? Everybody's in every business line. And then all that matters is what comes out the end. You you can play with the, you know, the knobs all you want. Maybe you make less profit on your hardware and more profit on your services. You give the software for free. You make the money on the dirt and the infrastructure play, but you can kinda play with the knobs. Is that a Simant background? Simant? Yeah. Simant. '91. David Friedberg (00:08:12): 1991. Speaker 3 (00:08:13): Okay. Simant. David Friedberg (00:08:15): I think there was an era when the Internet kind of said to everyone, you've gotta recast how your business operates, how you do commerce, how you create value, how you run your business, who's doing what, everything was digitized. And now we're kind of moving from this era of everything being digitized, which maybe it started in the sixties and the seventies with the computing revolution, but then all of that data got connected. So everyone had to become a digitally connected business. And the era of AI is sort of like this digital intelligence business, and there's just a radical reinvention. And it's it's actually quite amazing how few businesses are actually spending money with Anthropic today and how few people actively use, like, AI tools every day. It's like, call it 1% of the population, which explains why people remain pretty pessimistic because they haven't really understood it yet. But digital intelligence, I think, really creates a reinvention in every category of commerce. I work in agriculture, and, you know, every day we kind of observe ways that things have been done historically that make absolutely no sense from a first principles perspective in this era today. And it's almost like in every one of these categories, you know, when everyone kinda rethought every business category from the Internet, it's like everyone has to rethink every business category from a digital intelligence basis. You end up realizing pretty quickly that it's not just that the businesses are gonna have to evolve, but the industries are gonna get rewritten. It's no longer gonna be this delineation of the way the supply chain works today. It's no longer gonna be this delineation of where value sits in the value stack in any given industry that suddenly one company can compress the value stack or one company can be multi vertical. Every single one of these industries perhaps don't even exist anymore, that there's some recasting that happens. So, yeah, I mean, I think that's probably right. It's a wild wild west right now. Chamath Palihapitiya (00:10:17): You know what's not gonna be valuable? If you take your free cash flow and you spend it on buybacks David Friedberg (00:10:22): Well, it says a lot about where you you don't know where to invest. I mean, that's a good point about Nvidia too. Chamath Palihapitiya (00:10:26): That's exactly right. Yeah. That that's to tell. When you are like, I'm gonna do a buyback, what you're basically saying is I don't have a clue. Jason Calacanis (00:10:34): You either don't have the high road and the hoot spot to buy something, build something, which would be the other two categories to spend large amounts of money. Big criticism of Apple over the years, Tim Cook's stewardship of that firm was based largely on taking half the shares off the table. Massive buybacks, massive dividends. And if you look at Nvidia, the CFO, as CFOs are want to do, brought up stock buybacks. They authorized 150,000,000,000 in additional buybacks back in September before Jensen was ousted. Total buybacks, 235,000,000,000. CFO said, now's the time. This is a top priority. Chamath Palihapitiya (00:11:13): I'm not pointing directly at Nvidia, but most of the EPS gains for many years in the stock market was because of this gain. Either because rates were low or because you had profitable companies, they took their free cash flow and what did they do? They bought as much stock as they could, they fought back the dilution they were giving via stock grants to their employees, and they would increase the earnings per share, which is this game that everybody was playing. And they were mostly rewarded for it, and they were creating a natural bid to their stock and the stock just trended up higher. And so you as a stockholder knew that if your company was going in to buy some percentage of the float every year, you were guaranteed, mostly, a five to 10% increase in the stock every year and you can just kind of play this game. That game is over. And the reason is what you said, Jason. Because what it is is a tell. And the tell is the following. I don't know what to buy. I don't know what to build. I don't know what to invest in. Which is okay when disruption is low. But when disruption is high, you are going to be rewarded for acquisitions because that tells the existing investor you are trying to defend the moat that you have and build new moats. That is why when Nikesh does a $4,500,000,000 acquisition, the stock goes up. The market is voting. It doesn't matter what I think about it. What I saw was, wow, the market is saying do more of that. Jason Calacanis (00:12:45): Well, and there's also, Freeburg, the game on the field. During certain administrations, we are anti conglomerate. The feeling in America has been... And I hate that this always seems to go back to politics, but here we are, folks. The market wasn't allowed to do big m and a. That was off the table during Biden. Now it's back on the table for Trump. We got maybe six month window before it closes again if people assume we're gonna have Democrats in office in 2028. So you gotta play the game on the field as it were. The time to buy stock back, I think, is when your stock is severely undervalued, Chamath. Or when you find something, you have to look at what you're acquiring and say, is that company accretive enough and can we grow it enough? Do we have a strategy post acquisition for it to grow versus buying our own stock? In other words, if I buy this business, me plus that business being accelerated by whatever we have, you know, is that better than buying our stock? Chamath Palihapitiya (00:13:48): You said a word I don't agree with. Jason Calacanis (00:13:50): Okay. Chamath Palihapitiya (00:13:50): But I do think that's how a lot of people think about it, and the word that you use, which I don't like, is accretive, which is where the spreadsheet jockeys get into a room and try to figure out how it just adds to the value of the company today without thinking strategically how does it add to the value of the company over some number of years because it puts you into business lines that you're not in today. And I think you have to be in the latter category. And if you're beholden to an equity market that won't give you that permission, you're in a really complicated position because they're also not gonna reward you for buybacks, and they're not gonna reward you for dividends. What are you to do with this money? Except invest your way out of it. So this is the weird thing where the owners of these companies don't seem to get that there's a tsunami of disruption coming and that they're going to have to position themselves defensively and offensively by having a full stack solution. That's the only solution. There's no way around it. Jason Calacanis (00:14:47): And if you look historically at what's happened in our industry, last ten years, Apple has 10... Over the last ten years, Apple has 34 less shares. Alphabet, 12% less. Meta, 11%. And and and Meta's done some great acquisitions, but this is what people have chosen to do with their war chest. Now the other option, which I kinda like, is building out infrastructure. We talked about that. You know, if you're Meta, you're Amazon, you're obviously Alphabet, yeah, building out infrastructure and having the platform to create intelligence, that seems like a pretty wise investment as well. Yeah, Chamam, you think? Chamath Palihapitiya (00:15:25): Yeah, I mean, think the problem where these guys will fail is I don't think that they understand what the next generation application stack really needs to look like. That needs to be acquired. And the reason is it's just much more likely that small groups of people bleed off of these big companies or bleed off the frontier labs and go and build the things that are the new thing. They understand it better. And I think that that's okay. The point is you have to end up at the finish line with a solution. And I think it's actually simpler to say I'm gonna just acquire my way to it. And that's actually a safe harbor. Meaning if I was owning an equity and the CEO got on an investor call and said, you know what guys, disruption is coming. We're gonna have 10 mega companies. I intend to be one of them. I'm gonna acquire my way aggressively. I'm gonna use my cash flow to make sure that we don't lose. I would own more of that stock than all the other stocks where everybody's just singing from the same hymn book from years ago. Jason Calacanis (00:16:15): Yeah. That's not what I want. I don't see stocks and dividends. Anybody who can see the pace of technology, Freeberg, understands now is the time to to take some swings. What's your take on the anthropic IPO in the future of this company? 80,000,000,000 in revenue, a 100,000,000,000. They're obviously... The percentage growth is gonna slow down, but if we're at 2,000,000,000,000 now, they're trading 20 times the top line. Is this a company gonna just spend a couple years going sideways as it fills in that valuation? And and are you long, shorted? I know you didn't like that question last week, but I think people wanna know from us, are we bullish or bearish or, you know, kinda moderate on the future of Ant after the IPO today? Obviously, it's great that they got out, especially in the turmoil going on now. So points to them for getting out the door. I think that was a wise move. But what's your take on the company itself now, Freebird? David Friedberg (00:17:11): It's Jason Calacanis (00:17:11): been David Friedberg (00:17:13): growing, like, beyond their expectations internally from what I've heard from people. So seems like they're easily on track for getting to a 100,000,000,000 of revenue, and I think they've got a lot of runway. We talked about them being probably less than 1% market penetrated in terms of their addressable users that could spend over $100 a year. So lots of runway. They definitely have the best enterprise product out there, better than Google and better than OpenAI. And at this point, I think Google's done four model releases in the last few weeks. A lot of people are murmuring that Google seems to have hit the RSI runaway kind of loop. If Google's done it, then I think it's likely that Anthropic is there as well, and it's a matter of time. There's certainly a lot of hesitation around the safety concerns in this moment in time, so there's probably a lot more being held back than people realize. The models are probably advancing at a pace that is faster than the human evaluation that can be done to ensure its safety. So, you know, Google's indication, I think, would be positive for Anthropic having also achieved some whatever the definition is of RSI, which means that the models improve themselves. They get continuously better, and you end up on this effectively kind of exponential curve on... One way to think about this is intelligence per dollar. So if you've got this massively scaling units of intelligence per dollar spent underway, this kind of system that kind of improves itself and delivers more, certainly sky's the limit. The economy is the limit for a company like this. And I think, like I mentioned last week, I I heard they did a great job in the roadshow of convincing investors that there is almost this, like, true natural monopolistic advantage inherent in the business that the RSI and the cost as that that results, meaning that the cost per unit of intelligence goes down dramatically, makes it an insurmountable moat that other people can't catch up to. The biggest risk is open source. So the argument might be that, you know, maybe open source is a risk that could eat some part of the business, but I don't think it's gonna eat the whole business because at the end of the day, if they're gonna have the greatest intelligence per unit of dollars spent for the advanced applications, they're always gonna be the choice. So it'll probably be between them, OpenAI, Google at this point is sort of the the race. The question really is how much excess demand was there that didn't get met in the order books as they got filled out, and how many of the $100,000,000,000 of share purchases that have happened today are folks that would sell in a short period of time versus wanna be long term holders, and let's call it 50%. David Friedberg (00:19:54): So you probably got $50,000,000,000 of true float against probably some multiple of a $100,000,000,000 of demand at the current price. And so the question is that there's obviously some elasticity to that pricing as it moves up, the demand will go down and, you know, the market will tell us where where that limit is. But given the growth rate, given the potential, there's an amazing story to be told here that if you're willing to buy the stock and own it for five plus years, you could hold a multibagger from from here at this at this entry point. And that's true for them. It's true for SpaceX. It's true for potentially Google even at this moment. So there's a lot of upside as the, you know, kind of economy gets gets rewritten around Jason Calacanis (00:20:35): AI. Shammoth, when you see it go public at 2,000,000,000,000, roughly 30 times their run rate, What's your take on the future of this company specifically? You heard Freiberg talk about, well, on one side of the ledger, you do have open source and we're gonna talk about the administration or a contingent of the administration reportedly trying to stop open source and TAR and signal who is using open source. Pretty dark stuff. We'll get into it. But just as we hit day one on Thursday, we tape... You listen on Friday. What's your take here on the future of this company? Do you believe it's gonna beat the markets? Do you believe it's a juggernaut? Do you believe it's unstoppable? Or do you believe it... Hey. Maybe we might be sitting here in three or four years and it's still trying to fill in this valuation? Chamath Palihapitiya (00:21:23): I think it's definitely gonna be one of these 10 mega corps. They're doing all the right things. My biggest risk for that business, I think they've derisked now, is balance sheet and having enough money. So I don't know about the valuation per se. If you look at it, it's had the exact opposite trading behavior of SpaceX, which doesn't mean anything, meaning that the day one behavior is the exact opposite where SpaceX traded up and then kinda came off at the late day. This one kinda started up already and then has been coming down. So on a day one behavior, who knows, it's hard to intuit. So it is a great business. It's a great long term business. They're doing all the right things, meaning they're going dirt all the way up to application. All that makes sense, but I think the problem is that the valuation of Anthropic over the next three or four months has mostly nothing to do with Anthropic. If you look at other businesses that are making gobs of money, Nvidia is a perfect example. We've never seen a quarter like this ever, and the stock is down, and it's down meaningfully from the earnings day. Why is that? The market, when I look at this, I could change my mind, but right now what it looks like is it's trying to will itself to go up. And the problem with that is that inflation is sticky, all of these external conditions are not resolved including the war. I think people on balance feel skittish, they feel like this is run for a while. And I think the balance of trade is to de risk, not to add more risk. So I think you're gonna have all kinds of pockets of opportunities to buy stuff much cheaper. And then on top of that, what you're seeing, Jason, is this very weird behavior where there is just these crazy acute moments, just as we are taping. AMD is down 10 or 12% because somebody put out a short report that they were selling chips to China. I don't know if that's true or not, but 10, 12% trade in a trillion dollar company is ginormous vol. So I guess what I do know is that the ten year unfortunately keeps ticking up. That's bad for equities. Vol is elevated. Chamath Palihapitiya (00:23:44): That's bad for equities. And inflation is sticky. That's bad for equities. To the extent Kevin Worsch wanted to cut, I don't think he's got much room, and if anything, he's going to hike. That's bad for equities. There's a new sheriff in town in the Senate and in Congress. That creates uncertainty. That's bad for equities. Speaker 3 (00:24:08): Yeah. Chamath Palihapitiya (00:24:08): So... The wall of worry now is bonkers. Yeah. Maybe you can't be short. I don't know. Because, like, things could still rip in your face. But it's hard to justify how you get mega long here, what you're aiming for. What is the setup where the S and P goes to 8,000 from here versus 6,500? Jason Calacanis (00:24:28): I'll take a shot at it. Chamath Palihapitiya (00:24:29): I don't understand it. Jason Calacanis (00:24:30): I got... I'll take a shot at it because Chamath Palihapitiya (00:24:31): I'm Jason Calacanis (00:24:32): for long... Yeah. And I think this is the... I I think we're bouncing along the bottom independent of, like, a crazy black swan type event, COVID, etcetera. But that wall of worry that you very accurately described, you got rates, you got the debt, and you got inflation, you got wars, and you got political turnover, and the people coming in hate AI, big tech, capitalism. They're gonna be in the breakup big tech. We could be having a conversation in two or three years about, you know, let's say you're at the... If Chamath's thesis is correct, we're going into the mega corp, you know, AOC and Bernie Sanders and that constituent might say, yeah, we're not gonna allow that, and we're gonna go the opposite direction. We're gonna make you break up these companies. But the bull case that I see is we have only 10 or 20% of workers at the major companies using this technology. I invest in a 100 startups a year. I see how they use it. They use it all day long. They're using it, the average worker, five, six, seven hours a day. In other words, they're working with AI more than they're working with humans. What startups do is what the big companies eventually do. It takes two or three years. Startups embrace the cloud, and we work because it's more efficient. Then sure enough, the big companies will do that. So I'm watching that game on the field. I think we're gonna see the number of people in the organizations use it, you know, obviously go five, six, seven x because there are very small number of people in organizations using it. But more importantly, they're gonna go from using it for minutes a day or maybe an hour to using it for six, seven, eight hours a day. That's gonna be the primary use case at work, especially in in in corporate America. You're going to be working with agents and you're gonna be coordinating 10 virtual workers. That is the obvious thing that's gonna happen. Jason Calacanis (00:26:21): How do I know it's obvious? I see it inside my venture capital firm. I see it in my media company. I see it in 600 startups I've invested in. It's obvious. What that means is at least 500 times the token usage is what's gonna happen. Now where do those tokens come from becomes the question. Does it come from Anthropic? Does it come from SpaceX, open source, AWS? And I think the answer is all of those. So what we're gonna be sitting here in three or four years, people are gonna look at this moment in time very similar to the .com era where people said, I think this whole thing got overhyped. It's a bubble. People lost their objectivity for what's actually going on inside of corporate America and then how the the individuals are using it. Individuals, once they start using bot technology, once they understand how to make loops and to and to get the best out of these things, they don't stop. They don't stop. And their productivity goes through the roof like 10 x, 20 x. Chamath Palihapitiya (00:27:18): I don't understand how you can connect that with why stocks are bottomed. Jason Calacanis (00:27:23): Well, the the bottoming of the stocks I think is rates, inflation, the overall economy being this headwind, and the political turmoil. People think this thing is overheated. I think it's underappreciated. That... That's just my belief. I think if you look at products, management, and market, the market here is going to go 500 x. The demand for this product is like a 500 x lift. And then the management of these companies, putting aside Jensen getting ousted, is just incredible. Like, the products these teams are making are extraordinary. The headwind is regulation, banning it, breaking up companies, and and that... Those are all real possibilities, but you're not... The product velocity now and the extent that people are embracing this technology, Chamath, I've never seen in my career, and I've watched four decades of this now. Chamath Palihapitiya (00:28:16): You've convinced me, and I agree with you, that usage is mediocre and hasn't even approached what the full capability is. You've convinced me of that, but you have not convinced me why the stock market has bottomed. Like I said, I think that it's more likely that the ten year is at five and a half or 6% than it is at four and a half percent, that just has a mathematical implication to all equities. Jason Calacanis (00:28:39): I like the pushback. Imagine you're running a company, you've got 200 knowledge workers working there, and they're currently spending on a $20 a month plan to a $200 a month plan. They become five times more effective at their job, 10 times more effective at their job, and they say, may I go from the $200 plan to the $800 plan? And you say, absolutely. I have people coming to me in my own organization where I'm watching. Chamath Palihapitiya (00:29:04): You're saying revenue's gonna go up 50 x in this company? Five x to 50... Jason Calacanis (00:29:10): May you will be able to create more products and services and deliver them at a higher What product? Revenue and earnings Speaker 3 (00:29:18): per employee. Chamath Palihapitiya (00:29:19): Virtual products, you're saying? Not physical products. Because physical products still have input costs and those input costs are going So Jason Calacanis (00:29:25): I'm talking about software services, knowledge work, all of that is going to go from 500,000 to a million dollars an employee to 3 to 5,000,000, maybe even $10,000,000 per employee. And if you look at companies like Microsoft, they were at, I think, 900,000 an employee, and now they're trending towards 2,000,000 per employee. That trend is going to happen every three years. You're gonna see the amount of revenue per employee double, which means if the revenue per employee doubles, that efficiency level, that means you don't care if they spend 10 x on software because the software was only 10 of... Or five k of their overall cost, maybe five k of their overall cost a year. So if it goes to 25 k, you don't care. You're like, great. You're a superhero inside this organization. Chamath Palihapitiya (00:30:11): I hope you're right, but I think the market is in a show me state. Jason Calacanis (00:30:14): Sure. You have to prove it. Chamath Palihapitiya (00:30:15): I don't think they buy that. Jason Calacanis (00:30:16): You have to have cojones to be buying Nvidia right now at a 150, 160. I did. You gotta have cojones right now if you wanna buy some of these stocks, I think. Why do you say it's cojones? Chamath Palihapitiya (00:30:26): These are just mathematical risk adjusted EV decisions. It's not about, like, Jason Calacanis (00:30:30): No. If it was mathematical Chamath Palihapitiya (00:30:31): just gas pumping. Jason Calacanis (00:30:33): No. I I do think you have to have conviction Chamath Palihapitiya (00:30:35): because Jason Calacanis (00:30:36): you have to believe, and I don't think it's a mathematical certainty that it goes 500 x in consumption. That's my belief based on what I see. But I think a lot of people are doing the mathematical calculation just saying it's impossible to get there because they don't see just how transformative the technology is. That's what I'm looking at. How transformative is this? How much will people spend on it? I I I think it's being greatly underestimated. Friedberg, you've listening to this conversation. You have any input here on the the bull bear case here for the deployment of the technology? And what are you seeing inside of O'Halo? Are you seeing people 10 x their usage of the technology, and are you seeing the Chamath Palihapitiya (00:31:14): No. That's not the question. Of course. Asking him to Is revenue up 10 x? Jason Calacanis (00:31:18): Well, that was the second part. Are you seeing a multitude of consumption of tokens and a multitude of efficiency and revenue, or are you predicting that? Because it does take some investment to get to there. David Friedberg (00:31:29): I'll repeat what I've said in the past, which is I think the biggest opportunity for AI is on the frontier. That is where basically businesses and people can do things that humans and businesses have never been able to do before. And that means crafting solutions to really difficult engineering problems, to really difficult scientific problems, to previously untractable problems. And that's where I think that the frontier gets expanded by a 100 x or a thousand x versus AI being some, like, iteration on existing business processes to save money and make things more efficient, which is where people natively go because most people don't work on the frontier and they're not kinda pushing the envelope on what's possible. So their mind is always about, well, how does this change my job? And then you immediately get fearful about your job getting recast. And the truth is that's not the typical use case. The typical use case where we spend time and spend tokens is doing things that we've never done before, figuring out how to solve those really tough problems, how to do it quickly, how to do it efficiently. And that means humans generally can expand our frontiers, our horizons. That's what I kinda see. And and I still believe that. And I I do think that there's this... We all get caught up so much in the kind of reinvention of existing businesses and cost savings and nonsense. The return on dollars spent on tokens or whatever is some small number when you're doing it to produce the same stuff versus it being a much bigger number if you use it to produce things that humans have never produced before, like some new cancer drug Chamath Palihapitiya (00:33:07): or Well said. Some new That's really David Friedberg (00:33:10): well said. Material science. And so I don't think, like, that's really where people spend time and money or will as the capabilities are realized. Jason Calacanis (00:33:17): Yeah. There's two things there. Chamath Palihapitiya (00:33:18): That's a really smart way of saying it. That's absolutely true. You throw tokens at the existing you're just gonna what can you ROI? You know, 50%, 60%, two point? You might as throw it at new shit. Jason Calacanis (00:33:29): There's certainly incremental improvement. I think you you stated it well there, Freiberg. You can incrementally improve any job. Which is also why the job loss narrative is bull. It's also why the... David Friedberg (00:33:38): You know, everyone's gonna end up on welfare is bull It's also like we need the government to step in and provide UBI is bull Because, like, with any other kind of frontier moment in human history, it's the new jobs, the new industries get created before the old ones are irrelevant. Jason Calacanis (00:33:56): And therefore, there's a bridge. And the demand David Friedberg (00:33:58): and the pull forward that happens, meaning, like like, why did everyone... You know, at at one point, San Francisco was the second most populous city in The United States because everyone pulled forward to the frontier to the gold rush that was happening during that small number of years when that happened. And so, like, everyone got on the train and they all went out there because there's so much more value to be realized if you're in San Francisco than if you're working in a factory on the East Coast. And so that is why everyone kinda jumped forward. San Francisco kind of, I think, has historically been this really interesting sort of place where technology takes leaps forward. So people do come out here a lot, as we know, as we've all experienced many times. Jason Calacanis (00:34:37): Season, we call it. David Friedberg (00:34:37): You know? Yeah. The boom and bust moments here in San Francisco. Everyone rushes to the next kind of frontier, the next wave. And that happens before the old wave gets recast and and and shut down. It's why we don't see rising unemployment because there's so much new stuff happening so quickly that it is happening before... Because it's it's such a more attractive way to put time and money and people and dollars is in the new stuff before you go and try and reinvent the old stuff. I'd rather, like, go and invent a cure to cancer and go after a trillion dollar market to put a capitalist hat on this thing at the same time than to go and figure out how to optimize the supply chain of some manufacturing business at a factory in the East Coast. Speaker 3 (00:35:19): Right. David Friedberg (00:35:19): And as a result, there's much more demand for jobs, for capital in that new frontier then there is a loss of jobs in the old world. And so I think we've got it all pretty wrong or the media has this narrative quite wrong, and I think, you know, that's why we haven't seen the jobpocalypse. It's why we don't see everyone rushing for UBI. Now there is still a k shaped economy problem in the sense that not everyone is coming along and benefiting and riding that frontier wave. And that is a more kind of fundamental disjointedness that's happening in this moment that may actually prevent that frontier from emerging because of the negative backlash associated with what's happening. Because people see that frontier happening, they're not participating, they're not a part of it, they're not in it. Jason Calacanis (00:36:05): And so And the backlash is such a, you want me to build on this? David Friedberg (00:36:09): Mean, think to to everyone's view, it's just I'd rather see it shut down if I'm not in it. Jason Calacanis (00:36:14): Alright. I wanna go to the backlash here, Chamath. We we have two backlashes now, both from inside of... Both both emerging from the two political parties we're stuck with here in America. On the left, we talked about anti data center, which I think will soon be anti self driving and anti capitalism, wealth tax. You put that all in one bucket. Now on the right, we're having... And obviously, they have concerns, America First, on jobs and immigration. But here's a new one. White House is reportedly discussing a new open source disclosure rule to name and chain people using open source models. According to Reuters, companies could soon be forced to disclose on their websites if they're using Chinese open source AI models. No indication of how that would occur. Who's driving it? Apparently, treasury department Besant, who it seems is emerging as the leader of the China Hawk Group. This is all happening in real time, this civil war inside the White House. And then you have the White House OSTP, Michael Krasios, who I think you interviewed. Yeah? Jeffrey Berg? I don't know if you did it with Chamath. Reuters also reports that more people in DC are now calling for a tougher stance on China, giving the China hawks much more influence. Obviously, on the other side, you you you got, Alutnik and Trump himself. This year, Airbnb, DoorDash, and Cursor have all received letters from members of the house select committee on China. Because of their use of open models, Airbnb is saying, like, well, yeah, we use some open source models for our customer support. Breaking news, Chamath. I also am using a MacBook Pro, so I don't know if this podcast needs to have a label on it that we're producing it with Chinese hardware. Nothing has been decided in this, but it does seem like a wider crackdown. What are your thoughts on why now? Why now? Why is this leaking now? Jensen being ousted, Anthropic going public, the D cell movement, is this all coordinated or is it all just a very interesting amount of timing? Chamath Palihapitiya (00:38:21): Can I answer your first question which is Jason Calacanis (00:38:23): Sure? Chamath Palihapitiya (00:38:24): What will it do if you put a disclaimer on your website that you use an open source model? Nothing. It'll mean that you're not retarded. Jason Calacanis (00:38:32): So it's the not retarded logo, got it. Chamath Palihapitiya (00:38:34): Think about what the market wants. What does the market want? The market wants you to be smart. It wants you to be budget friendly. It wants you to be ROI oriented. It wants you to own your own technology. It does not want you to leak your alpha. Right? Jason Calacanis (00:38:52): AI sovereignty is what we bundled that under here on the podcast. Chamath Palihapitiya (00:38:55): That's what they want. Otherwise, why would they wanna own your stock? Because what is the alternative? For you to say, I don't really care about cost, I'll just take whatever model is available to me, I'll feed them all the traces and spans, they may learn how to compete against me by me using their own product, they tell me they're not gonna do it because of, you know, zero data retention policies, I trust them, it is what it is. The market will think you're an idiot. And why will they think you're an idiot? Because you would actually be an idiot. You would demonstrate that you have such a lack of technical awareness, that there is nobody even half technical on your team, that you have an inability to even recruit a consultant to help you explain what happens. Call Alex Karp. He'll send all kinds of FTEs, brilliant engineers to tell you what the risks are. And then you can make a decision. So if the decision is disclaim a thing that nobody's gonna see on your website or go out of business Speaker 3 (00:39:48): Yeah. Chamath Palihapitiya (00:39:49): I would disclaim the thing, not You go out of Jason Calacanis (00:39:52): could do a paratrade, one of your favorite things, of the companies that are smart and dumb. So maybe what this actually says is do the Chamath Palihapitiya (00:39:59): disclaimer, who cares? What it really speaks to is I think Washington needs some extremely technical people on the inside that explains the actual technical implications of some of these things. I think that the thing that was floated there was a trial balloon And I think if you go through the series of explanations of what it means and doesn't mean to use this stuff, you would come to the conclusion, well, wait a minute. To not say that you'd be using cheaper open source alternatives to retain your edge is more destructive to American companies. Is that what we actually want? The opposite disclaimer or that you don't use a model at all? So I I just think, like, these things aren't well understood. Friedberg, if you if you look at open sources, this shows a fundamental misunderstanding of how open source works. You're not buying it from the Chinese. It's on GitHub. It's on a website. You take it. You fork it. You put it on your own servers. You use it. You control it. There's no backdoors. And if there were, you would immediately have a thousand AIs pointing at the back door in it. It's kinda ridiculous. If it was, hey, I'm using Huawei technology, yeah, perhaps you could say, hey, this person's using something, you know, it's hardware based. But what were your thoughts on this very strange leak and why it's leaking now and why it's floated and why they're sending this to Airbnb? This seems like complete utter incompetence on whoever created this concept. What did they send to Airbnb? Earlier in the year, they sent a letter. It's coming out now in this Reuters report that this letter was sent. DoorDash got one and Cursor got one. So these are letters from the house select committee on China for American companies using open source models that originate in China. I would think DeepSeek and GLM five two, you know, etcetera. David Friedberg (00:41:46): Yeah. I think this continues to be the path of tyranny and any effort to try and curb open source, restrict open source, label open source, and give the government the tools to create effectively regulated monopolies through the anti open source efforts that are being undertaken, I think is gonna both limit human prosperity, limit individual liberty, create an Orwellian state, and I think fundamentally violates freedoms of speech. And I don't think that the origin of open source software should matter. Some people might disagree, but when you download open weight, source model from a repository, which is means a website where these things are stored and anyone can get a copy of them. You can then use it on your own server, your own computer, run it in the background, You can rewrite it. You can reuse it. And this was the basis of the Apache Foundation's tool that that created open source servers that allowed us to have the open Internet, the open browsers with the Mozilla Foundation and Firefox so that Netscape and Microsoft couldn't restrict people's access to the Internet, and we had an open source free version that people then ended up making many different forks or many different versions of. And what Google did with Android where they created open source operating software for mobile phones and found hardware companies that would make it the de facto installed software so that 70 of phones in the world today have unfettered access to the Internet. Whereas before with closed source software, the companies were going to try and control what apps you could put on your phone, what websites you could access, everything was monitored, everything was controlled. We're now facing the same decision point in open source AI as we did with all of those other prior technologies, and open source needs to be allowed and frankly encouraged to proliferate because the absence of that means that you're gonna have tyrannical groups. David Friedberg (00:43:59): Ultimately, they become tyrannical. They don't intend to be tyrannical. But every group that has the power to control what people see, what they're allowed to do, what they're allowed to access, how they're allowed to use things, what those specific weights are on, quote, unquote, facts of history, on, quote, unquote, proven science, on, quote, unquote, the truth. That is where people become tyrannical because they have an opinion where they think they're right. And if they have the power, they will act on that opinion, and it will ultimately give them tyrannical force over a broader population. That's what will happen if we limit the proliferation of open source software in AI. And I've said it before, I'll say it again, ultimately you can regulate, you can restrict, you can control, you can litigate, you can criminally penalize people for the outputs of what they use software to do. But to limit software, to label software, to make it a point of restriction and control creates a regulated monopoly for the closed source products that are out there that ultimately, if aligned with certain government agencies, gives those government agencies and those individual corporations power and control over you, the individual, what you're allowed to do, what you're not allowed to do in every business on earth, no longer operates freely. None of us should want that. I don't care what your political affiliation is. Chamath Palihapitiya (00:45:22): Would you David Friedberg (00:45:22): support Chamath Palihapitiya (00:45:23): a bunch of these cash rich public companies now standing up a bunch of open source model initiatives inside of their companies? Would you view that? David Friedberg (00:45:33): 100%. Yeah. Yeah. It's great. And I think Meta should... Think it's the play for Meta. Know, know that Meta's already doing it. Jason Calacanis (00:45:38): Google's already doing it. But they're not doing with Frontier. Meta's open are nerfed. Chamath Palihapitiya (00:45:45): They've been nerfed for a while. Speaker 3 (00:45:45): They're not not good. And by the way, I David Friedberg (00:45:47): think that one one point to note, the idea that you can quote label something Chinese. So in an Internet that's anonymous where people can post on Hugging Face a new open source model, the origin of that model may not need to be defined. So someone could just post a new model on Hugging Face. We all see it. It's the best open source model ever and it doesn't say what quote country it comes from. It was trained somewhere. It was built somewhere. Some set of engineers wrote this thing, posted it, and now it's available to the world. What do you do? How do you classify that? Why does the country of origin matter as much when it comes to open weight, open source when anyone, because of the open weight, open source nature of it, you can go in and It's a fundamental misunderstanding. Speaker 3 (00:46:31): Yeah. Jason Calacanis (00:46:31): These people don't understand how it works. David Friedberg (00:46:33): Can go in and they can test it. Anyone can look at it. You can see there's no there's no runtime software in there, just to be really clear about this. Open source OpenWeight is not a runtime application. You don't download it and run it on your computer. You have to have an environment in which you're running those open source OpenWeights against. So this isn't some like... There's some hidden tricks or tools inside of that software that can screw your computer over and hack you and take over. That's not the way this works. Technologically, people I think are deficient in that understanding. And then they're trying to make rules and regs around like, well, you don't know what the Chinese are doing. It's like, well, actually we do. We fundamentally concede. In an open source open rate model commit. You see everything. There is no there. Yes. It is not it is not an executable. It is a set of data. It is a piece of data, a chunk of data, a CSV file with a bunch of stuff in it. That's all it is. And you are deciding how you will then take that data, and you will use it in making your own applications and your own software. And you as a builder of software can decide how you're gonna use that. So in fact, this misunderstanding where people say, well, we don't really know is incorrect. We do know. And you can look and you can adjudicate and you can audit and you can test and you can do all the things you wanna do for safety and security screening. In fact, everyone faces liability if they don't Because if you then take something that does something that doesn't work, your product's gonna suck. You're gonna break the law. There's all sorts of bad things that could happen. It's your job to adjudicate whether the open source software does its job. That's your job, that's the way open source works. So to have the government come in and suddenly try and create their genius patch, their Band Aid idea of telling people to label stuff and not label stuff is as idiotic as I would expect it to be given the lack of engineering competence in the United States Congress. And that's not political, by the way. That's just fact. Jason Calacanis (00:48:20): Well, I mean, there is something political going on here, Chamath. If you look at when Jensen was on... Or we were at Jensen's conference and you look at the discussion we had and we were talking about Nemotron and then he started buying some open source assets, we now have emerging a series of events of the Chinese hawks and maybe the open source pro open source movement. All of this is happening in the same time period. There are no coincidences as I think a rubric many of us go by. I certainly do. I don't believe this is all a bunch of random coincidences. It's pretty clear Jensen was going to become the open source champion. His models, you mentioned, conglomerates from, you know, dirt to the application layer and beyond. He was building that. He was building that. He was suddenly deposed. Then you have this, we're gonna name and shame Airbnb and DoorDash for using open source. Okay? That's happening at the same time, and Ant is going bubbling at the same time. It feels like there is a giant civil war going on and a global conflict over open source versus frontier models and who's the champion for them? Chamath Palihapitiya (00:49:36): I don't think it's that nuanced. I think it's more about power and I think it's about control and I think it's about ideology. Speaker 3 (00:49:42): Okay. Unpack. Chamath Palihapitiya (00:49:43): Yeah. I think the ideology is whether you believe that this advantages the Chinese or not, and I think that there's a strong case to be made that it does, but there's a strong case to be made that it doesn't. And I think depending on how ideologically grounded you are in neoconservatism, you will bend in one way or the other. I think the reality is that there are things that, in hindsight, again hindsight's twenty twenty, that could have been done to create escape velocity for open source. It's really a shame when you look back on it. With all the money that the domestic American companies have, why did we not find a way to spend some of that money to get real escape velocity for a true American open source alternative? I think that's an important question worth asking. Speaker 3 (00:50:38): Why Chamath Palihapitiya (00:50:38): are the only open source alternatives Chinese that are any good? Why? Jason Calacanis (00:50:42): I think it might be. If you look at the talent, the talent had two really good options and and a third and fourth. Work on proprietary models and make hundreds of millions of dollars or billions of dollars personally for that talent. So the talent went to OpenAI. They left Google, which was not paying attention to AI, maybe to the extent they should have. And then you have these two major money grabs opportunity and all the talent flowed to them and then obviously to to X. Chamath Palihapitiya (00:51:12): But that's not true. That's not true because there are some really talented people at Meta. There's some really talented people David Friedberg (00:51:17): I'll tell Chamath Palihapitiya (00:51:17): you why. At Google. There's some really talented people in NVIDIA. The question is why couldn't we spend 50 or $100,000,000,000 of the trillion dollars of cash that we've spent Speaker 3 (00:51:25): Tragedy. And not had a Chamath Palihapitiya (00:51:26): not had a really good alternative. Jason Calacanis (00:51:28): Friedberg, why? Too much money to be made in the frontier models. It was the it was the incentive. Show me the incentive. I'll show you the outcome. David Friedberg (00:51:34): It's not just money to be made. It's capital to be raised. I think it was a lot easier over the last couple of years to raise capital as a private frontier model company than go out and say, I'm gonna build an open source AI company. You're not gonna be able to attract. So when you're like, hey, I'm gonna build a model. You wanna be a frontier lab. You get a bunch of great geniuses together, you're like, let's go build a new frontier lab. Your two business plans are open source. How are we gonna make money? I don't know. We'll figure it out later. For the good of humanity, I don't know. Or closed source, and we'll make money on the the standard business model. The venture money You're saying the VCs Chamath Palihapitiya (00:52:10): perverted the incentives. David Friedberg (00:52:12): Yeah. Capitalists in general. Right? So capitalists in general created the system where you could plow lots of money. Whereas for the Chinese, it's actually a very different game theory because for the Chinese, it's how do we disrupt American capitalism? How do we disrupt the American AI ecosystem? How do we create an alternative Chamath Palihapitiya (00:52:29): to David Friedberg (00:52:29): closed source where America controls the world? Blah blah blah. So for the Chinese, it's very rational for them to And when you're Jason Calacanis (00:52:34): buffering David Friedberg (00:52:35): open weight model. Chamath Palihapitiya (00:52:36): Wait. Hold on a second. But why isn't that the same game theory if you're Nvidia or Facebook or Amazon in the sense that... You know what you mean about that? Speaker 3 (00:52:47): I think you're exactly right, I think we've been on the precipice of David Friedberg (00:52:49): that Speaker 3 (00:52:49): being the case, and I think it all got blown up in the last week. Jason Calacanis (00:52:52): Yeah. That's why Nvidia bought all of these open source resources. So that actually tracks with your original theory. When you're I think behind Speaker 3 (00:53:00): everyone's behind. Jason Calacanis (00:53:00): When you're behind, you go open source to try to kill the existing codified solution. That's what Android did to to iOS and now that's what NVIDIA was doing. They literally bought all those open source products. Chamath Palihapitiya (00:53:14): Yeah. But none none of those things were about models themselves. They're about hosting models. That's all fine and good. But you could have given more compute to Nemotron. Jason Calacanis (00:53:20): Which he was doing. He was starting that process. Chamath Palihapitiya (00:53:23): If you look at the checkpoints on Nemotron, they're unfortunately behind. So for whatever reason, whether it was compute constrained or talent constrained, the models are good, but they're not great. And the GLM models and and the other Chinese variants are better than good, also still not great. And, obviously, the frontier ones are great. My question is that boundary condition that both of you just talked about was true a year ago. It was true a year before that. Why wasn't the money spent then? Why didn't we score... If we, meaning, if I was too married David Friedberg (00:53:53): to be made with... Like, there's just been too much capital. Look. I mean, look what happened today. Today, the largest fundraise in human history just took place. Anthropic in a day raised 85,000,000 I Chamath Palihapitiya (00:54:03): know for a private company. I'm saying Follow the money. If I was running AI at Meta, I would say, you know what I'm gonna do? I'm gonna scorch the earth right now. Two years ago, I would have said this. If I'm gonna spend $1,000,000,000 on android taluk and a billion... Yeah. Jason Calacanis (00:54:16): They tried. Chamath Palihapitiya (00:54:17): They tried. I think you're right, and I think they did, and I think that they just didn't get there. Jason Calacanis (00:54:21): Right. And invade It's just that simple like that Chamath Palihapitiya (00:54:24): they just... The models suck? Jason Calacanis (00:54:25): Well, no. They're getting there. It takes time. They're they're two years behind. You you explained the theory perfectly, but not getting to the conclusion. Yes. People realized they were behind, and there was an opportunity there. And then Jensen bought, and this might have been his downfall. He bought Hugging Face, and he bought Poolside. He spent 20,000,000,000 on those two companies. So he said, yes. We're gonna double down, triple down on this. And Meta and Zuck said, yes. We're gonna do it as well. Now you said they were nerfed. Yeah. Okay. So they didn't get it done, but the intentionality was there and it's in process. And I believe that's at the core of why Jensen was Do have a different David Friedberg (00:54:59): theory, Chamath Palihapitiya (00:54:59): Timoth? I kind of have like a like a working version of one. Jason Calacanis (00:55:04): Well, let's workshop it. Go ahead. That's why we're here. Workshop it. No mistakes here. No bad ideas. Chamath Palihapitiya (00:55:09): I'm really at a loss because if it were me, I would have stuck a... Again, hindsight's twenty twenty. This is where, like, you know, we we did talk about this, but if I was an entrant at that scale with that balance sheet two years ago and I was far behind, it made a lot of sense to commoditize the model there if you were a non model native company. Yes. Right? You make the money in the apps, you make the money in the chips, you make the money in the LPS, you don't need to make money at the model. That seems obvious. So you pay a bunch of money and you try to spin a model and what do you learn? It's much, much harder than you think. And I guess I still struggle with isn't it just a money and a recruiting problem because what you guys are saying is VCs created incentive to come and say we're gonna build a closed frontier model or we're gonna build an app company and you could raise lots of money. So I don't know, I'm still like, the only theory I had was they took the eye off the ball they just got distracted with, hold on. Just let me finish. They just got distracted with all this other bull Jason Calacanis (00:56:13): and it's like, just finish your breakfast. You're correct, dad. In hindsight, and if you were a great strategist, you would just carve out 25,000,000,000 and made an open source project at Microsoft, at Google, or Apple, whoever's behind Meta, and and just make that your way to be... Make that your strategy for open source. I think two companies were exactly pursuing that, Zuck and Jensen. Now if you look, though, at the money and the incentives, what is the average person, Freiburg? What is the average person going to work at OpenAI who's a computer scientist or going to work at Anthropic, what is their pay package like? These people got offered pay packages. I was talking to somebody whose kid out of school got offered $10,000,000 a year in equity at Quad over four years. This is a really good computer science undergraduate who got a $40,000,000 package. Like, this has never existed in the history of of our industry. And then remember that time when Zuck was trying to buy somebody for a billion dollar pay package? Remember he went crazy? Chamath Palihapitiya (00:57:20): And the talent. Jason Calacanis (00:57:21): Yeah. Yeah. So they kinda jumped the fence. In that environment, Freiberg's exactly right. How does an open source company compete? They don't have the the dollars. They don't have the market cap to buy all that talent. But I think now we're we're probably gonna see Muse Spark, Nemotron, and then Reflections Beam. Like, those three are just behind Kimi, Deepseap, GLM five three, and then you got this weird mistrial out in France, which is like kinda interesting. The French one is right between the three trailing US ones and the three leading Chinese ones. We'll sit here in a year, and The US ones will be somewhere in between Deepsea and Kymi. Chamath Palihapitiya (00:57:58): Let me give you a different theory. It could happen now, and it could happen now because you're seeing almost this schism where these leading edge folks are kind of splintering. You guys saw that there was this thing written called misanthropic, right? Was was that like a Yes. With like a bunch of signatories and it was a bunch of folks that are sort of like EA people that had this schism. There could be a version of this where really smart folks now at these model companies who have made a lot of money now start to leave and they're like, you know, it's probably important to have some open source alternatives and so maybe they go and work on this stuff in a more nonprofit way because they have that money and they have this predilection already to being a little bit less capitalist. So that that could happen. Jason Calacanis (00:58:50): No. I think it's, like, a very interesting insight. And, obviously, you might... Many people in the audience probably saw the New York Times story here Dostoev Moskovitz early investments in Anthropic and OpenAI, big profile in him being uneasy about all this stuff. And I do think it's a nonzero chance a lot of these people will get together and build the alternative to what their proprietary models are putting out there. Let's end on NVIDIA. They've shed $2,000,000,000,000 in market cap in the last thirty days, down 35%. Two weeks since Jensen got shivved, it's down 20%. CFO and acting CEO Colette Kress didn't exactly inspire confidence during her first earnings call, but the earnings call was extraordinary. NVIDIA beat the top and bottom line. Revenue and net profit both doubled. Speaker 3 (00:59:40): Yes. Jason Calacanis (00:59:41): Doubled on already absurdly large numbers to a 116,000,000,000 and 65,000,000,000 respectively. Let that sink in. $65,000,000,000 in profit. Most companies trading in the stock market are not worth what Nvidia just made in profits this quarter. Stock, even after all that, dropped. Internet's having a field day with this. Here's here's your viral moment of crest on Mad Money with Jim Kramer. Speaker 3 (01:00:10): Jensen resigned from his position with NVIDIA, and the board accepted his resignation. The board has determined that they will not reinstate Jensen, and that is our statement and what the board has decided to do. Jason Calacanis (01:00:29): Internet, obviously, having a field day with this. Everybody is very pro Jensen, it seems, except for the board of this company. And no more splashy cashy is one of the messages here. Crest said they're gonna be more disciplined and selective, quote, more disciplined and selective on new neo cloud financing. Obviously, Jensen loved to invest in startups and and companies that were building the future. So the Bank of Jensen is closed, and she ended the h 200 sales to China. H 200 is obviously a more powerful version of the flagship. H 100 NVIDIA is allowed to sell h two hundreds to China. Commerce approved those for sale in January. I know this is all very complicated, folks, but this is an elective move according to the news. She's just saying, you know what? We don't want anything to do with this anymore. The the most powerful GPUs, Rubin and Blackwell, are obviously banned from being sold. China and the board repeatedly is saying Jensen cannot come back. Jensen said he would humbly like to come back, and there seems to be a group of people mounting his comeback to a certain extent. We'll see. In related news, NVIDIA is facing mounting legal issues. You predicted this. Chamath, SEC probe, class action lawsuits, both of those are focused on misleading statements potentially that were given to investors. DOJ is obviously continuing their investigation on the relationship with mega speed. The other big problem here is talent. You got a bunch of employees signing petitions, and Travis, no stranger Travis Kalinick, friend of the pod, no stranger to getting ousted, said the employees should do a walkout. Here's the clip with other bestie, Antonio Gracias. Lot of leather jackets on that on that dais in support of Jensen. The leather jacket protest continues. Speaker 3 (01:02:20): Now I obviously experienced this. I went through six months of ops before I went down. I held on as tight as I could, but eventually, I just couldn't. But any wisdom I can provide about that experience to Jensen or the team of people around him that I think are ultimately gonna lean in to fight to get him back because it matters so much, that... That's what I'm bringing to the party. The people said, hey. Hey. Is the petition, like, is that good? I'm like, dude, I had... I sent I sent Jensen, like, you know, the the screenshot of the stories when there was a petition for me. I'm like, petition good. Yeah. Walk out better. Right. Right. I didn't get a response. Yeah. Yeah. But I will I will show up to NVIDIA HQ with a wafer encrusted megaphone with tons of NVIDIA employees who wanna walk out, tons of CEOs, founders who believe in what I'm... What I what I just said, who are believers and don't want an AI overlord, and let's fight for what's right. And give it a shot. I... You can't... We don't have control of the outcome. Alright, man. But let's fight. Jason Calacanis (01:03:28): There it is. The industry lining up behind Jensen. Leather jackets up for Jensen. Jackets for Jensen. Next week, I expect everybody to show up in a leather jacket. Your thoughts here, Chamath, on the NVIDIA situation, and then we'll go to you, Friedberg. Chamath Palihapitiya (01:03:44): I think it's very complicated still to have a definitive opinion beyond the following. He was an exceptional legendary founder and CEO. He built the most important company in the most important industry in the world. Speaker 3 (01:03:59): He is Chamath Palihapitiya (01:04:01): biblically good. He also made a bunch of mistakes. The extent of those are still not known, and the problem is that you can't reinstate somebody like that into a public company with that much valuation that has been destroyed because it just completely throws governance out the window and it puts everybody at risk, everybody, including the company. So that's why it's not gonna happen and that's why they have to say that. And the unfortunate part is I think Colette is excellent and she's put in the awkward situation of having to be the message bearer of that. And so then you get the internet trolls that run around and give her a hard time and it's not her hard time, she's just a messenger. Speaker 3 (01:04:41): He Chamath Palihapitiya (01:04:42): is exceptional, that company is exceptional, but they are in a very unique moment where you cannot make any decision. I repeat, you can't make a decision until all of the congressional hearings are done. I mean, I have to stress, these guys have to navigate three different committees on December first and second. The Select Committee, the House Intelligence Committee, another committee. Do you think that that's easy? Does anybody think that what they're gonna be spending time with was pillow talk? So everybody needs to just calm down. He's gonna be really focused for December 1, and hopefully what comes out of that is there is nothing to see here. But I can guarantee you that if there's one even shred of there is something to see here, it's going to get leaked and there's going to be another wave because the stakes are too high. Right? Because you have an Internet mob versus the China Hawks versus... Who knows all of the different players that are at that play here? I don't know. That's why, as I said before, this is so in the too hard bucket. And so the more people blather on about superficial bullshit, the more you actually make it harder for him to come back. Speaker 3 (01:05:58): Okay. Chamath Palihapitiya (01:05:58): So I would shut the up and I would let him prepare and I would hopefully have him nail this and prove categorically that there's nothing to see here. Then you can actually start to have a structured conversation. But all these ding a ling bottoms up protest betrays a fundamental understanding of what is happening, which is congressional closed door hearings on the record, which could result in some really slippery slope stuff for him, for other people, the company. And nobody should have an opinion until that's over. Jason Calacanis (01:06:33): Okay. Freeberg, what's your opinion? Given Chamath is just admonish everyone for not having one, I'm curious what's your take here and what's your... What what are your feelings about the situation and the founder being ousted and and etcetera. Speaker 3 (01:06:46): Yeah. Jason Calacanis (01:06:47): Look, what's David Friedberg (01:06:47): happened's happened and can't go back and change the fact that he resigned. We don't know whether he resigned over concerns about selling chips to China, whether there was espionage, whether it was based on affairs and violations of codes of conduct. We just don't know. The company has not put out a definitive statement. Jensen has not put out any factual statements. The government has not put out any factual statements or declarations. There has been no civil or criminal case that's been filed by the government against the company or by any of the government regulators against the company. So we are all just reacting to media that we always say all the time lies to us, so I don't know what to believe. Speaker 3 (01:07:32): I don't know what's true or what's not true. I... There have been no David Friedberg (01:07:36): official statements from the company, from the government, from the regulators, from Jensen regarding any of the facts of the matter. Jason Calacanis (01:07:44): Well, there are I don't know what there's a report. We... Well, there are some things we know. There there was a report that Rubio rewrote. We know that there was Again, Jason, that Speaker 3 (01:07:55): is media report. David Friedberg (01:07:57): What you were referencing is something that Media reports. Speaker 3 (01:07:59): Some reporter said. Jason Calacanis (01:08:01): Yeah. Yeah. I I said I'm not gonna Speaker 3 (01:08:02): react I'm not gonna react Jason Calacanis (01:08:03): to David Friedberg (01:08:04): media anymore. Speaker 3 (01:08:05): Okay. David Friedberg (01:08:06): So I don't know what's true. I just don't... Like, all I know is all I know is that the actual people, the first parties that matter here, government regulators, government agencies, Jensen and the company, none of them have provided any official data or evidence or proclamations. Chamath Palihapitiya (01:08:19): Exactly. David Friedberg (01:08:20): Everything has come through Chamath Palihapitiya (01:08:21): media, David Friedberg (01:08:22): and I don't believe any of it. So I don't know I what to Chamath Palihapitiya (01:08:25): agree with Friedberg. Everybody, if you want him back, to quote Dave Chappelle, you could just shut the up. Jason Calacanis (01:08:32): I I will not give my opinion even though that's your best advice, Tramont. I I actually think, you know, just looking at the totality of the reports and triangulating on them and knowing what I know about salespeople, this is a case of... And there have been reports of these servers and cloud providers giving a backdoor into some of this hardware to Chinese companies. This has always happened where sales of items get flipped and then provided to people. I bet you this is what's gonna come out. A bunch of rogue salespeople, you know, in a gray area, resellers on resellers, selling this, and it Chamath Palihapitiya (01:09:13): will wind up. You should be called as a special witness in December 1. Nauseamaz. Maybe like what you do. Nauseamaz could come. But this is this is a Jason Calacanis (01:09:23): podcast, Jermaz. What is the role of podcast in the world? Chamath Palihapitiya (01:09:28): You can look in your dirty paper brown bag of junk and see the future. Awesome. Don't know jack So I have chosen to shut the fuck up. Jason Calacanis (01:09:37): That's a great choice. I have chosen to speculate on a podcast, which is my right as an American. If we've learned anything over these years is that the goal of podcast is to speculate. Chamath Palihapitiya (01:09:48): You know what I've realized in the last few weeks? Speaker 3 (01:09:50): What if Chamath Palihapitiya (01:09:50): you're right? Here go. Here we go. I think it's really important to try to be right. Speaker 3 (01:09:59): Mhmm. Chamath Palihapitiya (01:09:59): And when you know, you say. And when you don't, you can guess. And when there's so much on every single side, it's just what Friedrich said, where it's so clear that you can't know, it's sometimes just better to just Jason Calacanis (01:10:17): keep your mouth shut. You you talked about that two weeks ago with the midterms. When when you talked about it on the midterms and you said, Mea culpa, hey. Listen. I got it all wrong. People were really excited about that money. Chamath Palihapitiya (01:10:27): Got it wrong. Speaker 3 (01:10:29): Yeah. Chamath Palihapitiya (01:10:29): I got it totally wrong. But this is a thing where I have no opinion because it's like the moment of truth has not come yet. Speaker 3 (01:10:36): Mhmm. Chamath Palihapitiya (01:10:36): It is a few weeks away. Speaker 3 (01:10:38): Yep. Chamath Palihapitiya (01:10:39): And so what everybody else is persecuting is online virtue signaling nonsense. Jason Calacanis (01:10:43): More speculation. Yeah. And people are speculating. It doesn't have to be virtue signaling. What Friedrich said... It is virtue signaling. Chamath Palihapitiya (01:10:49): What Friedrich said is totally true. Nobody who's in the middle of it has said anything on the record. Nobody. Nobody. And so what it is is just a bunch of goofballs on the sidelines with opinions. Jason Calacanis (01:11:01): Yep. Yes. What you just Speaker 3 (01:11:04): described Jason Calacanis (01:11:05): is what everybody in the industry is doing, speculating. Yes. And that is Not me. Chamath Palihapitiya (01:11:09): I'm not speculating. Jason Calacanis (01:11:10): Listen. I wish I'm the best. Speaker 3 (01:11:12): I hope not David Friedberg (01:11:13): I'll tell you what I'm speculating on. I'm speculating on the fact that there's a large scale unconscious effort to monopolize artificial intelligence, the greatest technology advancement in human history. Speaker 3 (01:11:23): Okay. David Friedberg (01:11:23): And if we don't all fight for the open source movement and we don't all fight for competition, the last thing any of us want, and I think Democrats, Republicans, regardless of your political affiliation or where you sit in the k shaped economy, no one should want an oligarchy or oligopoly when it comes to AI. This technology is going to evolve. It is going to move forward. You are not gonna stop it. So try and keep it from becoming an oligopoly by allowing the proliferation. Here's the numerical receipts. Chamath Palihapitiya (01:11:56): Today, as we speak right now, now it's eleven forty two. Okay? The S and P 500 is down 1.39%. The VIX, our vol index, is up 10.83%. Ant is up from the IPO price 18 percent. You know what the only other company that's ripping right now is? Palo Alto Networks up 5.86% because they made a huge bet on open source. Freeburg is right. And the market will reward you. And I think what the market is saying is it is philosophically right and it is mathematically right, and I think that that's a really important signal on a day like today. Jason Calacanis (01:12:38): Today is a bloodbath out there, folks. Chamath Palihapitiya (01:12:40): Here's your Oh, go ahead, Nostraconis. Look into your paper brown bag of This Jason Calacanis (01:12:45): is more looking at the market heat map, though. That would be This is actually a fact. Speaker 3 (01:12:49): Can't wait to see I can't wait to see it, by the Jason Calacanis (01:12:52): this podcast is number one is because we speculate. Just because you got burned because you touched the stove, and I told you, that stove's hot, Shemab. Be careful. Be careful if you get so close to politics. You burnt your hand on the stove, and now all of a sudden you're gonna be judicious. The ratings are gonna tank. Chamath Palihapitiya (01:13:09): That's not what I'm saying. I take speculative bets all the time. What I'm saying is I'm risk off because what I'm telling you right now is the balance of motion in this market is down, it's not up. I'm also smart enough to look at a market and say, what is it rewarding? It's rewarding the creation of one of 10 mega corps. So we're gonna have a mag 10. We know what the mag four or five. These guys make it a mag six. Okay? So they rewarded that today in a market where vol is up, people are scared, and, you know, they're mostly risked off, and they reward what Friedberg said, which is it is very important to now make big bets, use your cash flow, use your stock, get in the game, put on the pads, get on the field, hit somebody. I'm happy to observe that. I think it's wonderful for Nikesh. It's a brilliant thing that he did. He probably has a really good chance of being a trillion dollar business, you know, and building this thing over the next decade. Amazing. I wish him all the best. I am principally risk off because I think there's a lot yet to come. I'll put it in three buckets. Economically, there's a lot yet to come. Specifically around AI trade, we don't even know what the first order effects are because we haven't gone through these congressional hearings. Once you get past those hearings, then you're gonna figure out what happens when the thread gets pulled out of the Bank of NVIDIA. We don't know what the implications are for the bank of NVIDIA being closed. And I think the tenor of how closed or open it is will be more obvious after those congressional hearings, not before and not during. And then the third thing is we're in a situation where everything seems to be going down. We're getting cyber hacked every day. Everything is shut down. We're being told to prep for three days, and you don't think the balance of risks are down? Your paper bag tells you shit's gonna go up? Jason Calacanis (01:14:55): Oh, no. It's not a paper bag. I'm looking at the fundamentals in... Of your company. Speaker 3 (01:14:59): Plastic trash bag tells you that things are going up? Jason Calacanis (01:15:02): No. No. No. No. It's a Louis Vuitton book. Chamath Palihapitiya (01:15:04): What does Nostradamus use to look at the future? Speaker 3 (01:15:07): I see all Chamath Palihapitiya (01:15:07): of this. Jason Calacanis (01:15:08): I'll you. I'll you. I'll give you the three things. Products products, management, and market. That's how I make my bets. No. Speaker 3 (01:15:15): In his flossing bag. No. You have a dirty Chamath Palihapitiya (01:15:18): brown lunch bag. David Friedberg (01:15:20): Have a brown bag. Shaken? Speaker 3 (01:15:22): Bag. No. Jason Calacanis (01:15:24): I I... Trust me. Nostraconis sees the future, and the future is who's shipping great products. Chamath Palihapitiya (01:15:28): You don't see Jason Calacanis (01:15:29): I see Chamath Palihapitiya (01:15:30): You don't see I Jason Calacanis (01:15:32): I NVIDIA one fifty knot, not in that device. Chamath Palihapitiya (01:15:35): I can't Jason Calacanis (01:15:35): wait to see you Vegas. Vegas, everybody. Speaker 3 (01:15:37): Love you, besties. Jason Calacanis (01:15:39): Bye bye. Love you, boys. Speaker 3 (01:15:40): Bye bye. We should all just get a room and just have one big huge orgy because they're all just useless. It's like this, like, sexual tension that they just need to release somehow. Let

本集资源与出处