饮食TBPN：2025年11月4日

我们正在硅谷进行现场直播。我们正和一些顶尖人物在一起。红杉资本有突发新闻，他们迎来了一批新老板，新的掌权者。

We are live from Silicon Valley. We're hanging out with some absolute dogs. We have some breaking news out of Sequoia Capital. They have a new set of bosses, new top dogs.

掌权者？什么？红杉资本新一代的守护者。

Top dog. What? A new gen generation of Sequoia Stewards.

我觉得Roloff可能最近才接触赛车运动，他现在就像在说，我需要全职投入。

I think Roloff probably just recently got into racing cars, and he's just like, I need to go full time.

我必须全力以赴。我

I need to go all in this. I

必须 我必须

gotta I gotta

我找到了真正的使命。兄弟直接说风险投资是零回报风险，然后交出了钥匙。这个时机很有意思。

I found my real calling. Bro went out and said venture capital is return free risk and then handed over the keys. The timing is interesting.

Anthropic成为API销售的新王者，以下是大量新财务数据和预测。今年夏天Anthropic将其最乐观的增长预期上调了约13%至28%，预计未来三年收入将达700亿美元。

Anthropic is the new king of API sales, boatload of new financials and projections below. Anthropic this summer hiked its most optimistic growth forecast by roughly 13% to 28% over the next three years and is projecting as much as 70,000,000,000 in revenue in 2028.

是的。这里值得注意的是他们正在预测现金流。

Yeah. The notable thing here is they're projecting cash flow.

是的。2027年实现盈利。

Yeah. Profitable in 2027.

2028年170亿美元的现金流。这在NVIDIA以外的AI公司中可不常见。

17,000,000,000 in cash flow in 2028. Not something you see from AI players outside of NVIDIA.

还记得Dario那个观点吗？每个模型本身都能产生正向现金流，但你必须不断以10倍规模扩张。

Remember the whole Dario idea of, like, each model individually is cash flow positive, but you keep having to scale up by 10 x?

要在公司层面真正实现正向现金流，我想他们可能不得不放弃下一次训练计划。

To actually be cash flow positive at the company level, I have to imagine they would have to not do the next training.

OpenAI的计划是投入1150亿美元，到2030年实现盈利。Anthropic的计划是投入60亿美元，2027年盈利。我们会好奇哪种模式更有效。2019年时人们都说英伟达股票被严重高估——我记得那是在以太坊暴涨之前，当时英伟达市值才1000亿美元。

OpenAI's plan, spend a 115,000,000,000 to then become profitable in 2030. Anthropics plan is spent 6,000,000,000 to then become profitable in 2027. We'll be curious to see what works best. In 2019, people saying NVIDIA stock is extremely overvalued. This was pre the Ethereum pump, I believe, and NVIDIA's market cap was a $100,000,000,000.

2021年市值从1000亿涨到5000亿。而到了2025年，现在我们看到市值已经达到5万亿。这真是

Was at 100,000,000,000, then at 500,000,000,000 in 2021. And then in 2025, we now see that the market cap is 5,000,000,000,000. And this was

看到这张照片浮出水面真是超现实。我很惊讶黄仁勋会这么做，还允许自己被拍到。但这是2024年6月4日的事。没错，他在一场会议上给一位女士的衬衫签名，而英伟达股价涨了85%。

pretty surreal to see this image surface. I'm surprised Jensen did this and allowed himself to be photographed, doing it. But this was 06/04/2024. Yeah. He, was signing a woman's shirt at a conference and, NVIDIA is up 85%.

从历史上看，这些一直是

Historically, these have been

是啊，太疯狂了。

Yeah, it's crazy.

顶级信号，但我们还是走到了这一步。

Top signals, but here we are.

黄仁勋一直是个摇滚明星。他不是在巅峰时才穿上那件皮衣的，而是天生就穿着它。

Jensen has always been a rock star. It's not like he put on the jacket as at the top. Was born in it.

韩国炸鸡股票正在暴涨。等等。我做空韩国炸鸡公司导致财务破产。我失去了一切，不知道该如何继续。这看起来像是做空韩国炸鸡公司亏了4到17.2万美元。

Korean fried chicken stocks were pumping. Wait. I am financially ruined short selling Korean fried chicken companies. I have lost everything, and I'm not sure how to continue. That seems 4 to $172,000 short selling Korean fried chicken companies.

我的想法是：随着韩国人口减少，以及最近健康饮食和纯素食主义的趋势，这里存在机会。但现在一切都完了。

My idea was with the declining Korean population and the recent trend towards healthier eating and vegan diets, there's an opportunity here. Everything is gone.

关于ChatGPT的发布对工作岗位替代的影响有多大，目前存在一些争议。我们一直在反复讨论这个问题。我的意思是，他们当然希望达到那种程度，但目前人们还没有那样使用它。在大多数情况下，它更像是，

There's some debate over how impactful the ChatGPT launch has been on, job displacement? We keep going back and forth on this. I mean, they certainly want it to get there, but it, it, that's not how people use it just yet. Like for the most part, it is like,

我认为值得注意的是，当我想到ChatGPT时，关于工作岗位替代的辩论和讨论。是的。比如我看我们公司，尽管涌现了大量新工具，我们仍然需要招聘视频编辑。我们仍然想雇佣写手。有很多这样的不同角色，对我来说这张图表一直显示的是利率飙升。

I think the notable, I think, I think something notable about like when I think of the ChatGPT, like jobs displacement debate and discourse. Yeah. Like I look at our company, which is like despite a bunch of new tools popping up, we still need to hire video editors. We still want to hire writers. There's like a bunch of these different roles that this this chart to me has always been about interest rates going through the roof.

那那那正是它几乎完美契合ChatGPT曲线的地方。如果看看标普500的增长，几乎所有的增长都是由AI驱动的。

That that that's what it just happened to almost perfectly coincide with the chat GPT line. If look at the growth of the S and P 500, like almost all the growth is driven by AI.

要是一年前发生过什么重大事件影响了利率和招聘趋势就好了。所以就有了围绕埃隆在X公司所作所为的文化趋势，这让很多CEO意识到他们可以更精简运营。如果AI没有替代人力，利率影响就很大。但如果CEO们认为AI将要或五年内会替代人力，那么很多招聘的原因并不是因为今天需要完成工作，而是因为，

If only there were some major event a year before that was important and affected interest rates and hiring trends. And so there's, like, the the cultural trends around the what what Elon did at x that that just kind of, like, woke a lot of CEOs up to this idea that they could run leaner. The interest rate is huge if AI isn't displacing. But if CEOs think that AI is gonna displace people or AI will be in five years, well, a lot of the reason why you hire someone isn't because you need the you need the job done today. It's because, like

我认为每位CEO，即使没有AI叙事，也都想要一个AI叙事。所以如果要裁员，不如说：嘿，我们从AI工具中获得了大量效率提升，这就是我们裁员的原因。

I think that every CEO, even if they don't have an AI narrative, wants an AI narrative. So if you have to do layoffs, why not say, hey. We're getting a bunch of efficiency out of AI tools, and that's why we're doing layoffs.

是的。迈克尔·伯里刚刚提前11天提交了13F文件。他从未这样做过。大规模看跌期权，他正在做空Palantir和英伟达。

Yeah. Michael Burry has just filed his 13F eleven days early. He has never done this before. Massive put, he's going short Palantir and Nvidia.

我认为总的来说，人们持续误解了名义价值与实际规模的关系。是的。实际上

I think in general, people continuously misunderstand the sort of, like, notional value versus the size of Yeah. The actual

是啊，是啊，让巴里大展身手吧。

Yeah. Yeah. Let Burry cook.

我在想巴里是否希望自己在《大空头》中本色出演

I wonder if Burry wish he played himself in the big short

嗯。

Mhmm.

这样的话，所有那些梗图如果他知道每天会被分享成百上千次。接着说Palantir的财报，第三季度他们实现了11.8亿美元收入，同比增长63%。美国市场收入8.83亿美元，同比增长77%。美国商业收入3.97亿，同比增长121%。

So that all the meme if if he knew all the meme images would be shared hundreds, thousands of times a day. So that's more on Palantir earnings. Q three, they did 1,180,000,000 of revenue, up 63% year over year. 883,000,000 of US revenue, up 77% year over year. 397 US commercial revenue up 121% year over year.

业绩这么好第二天却跌了8%，说明市场有点不稳。

Beating like this and then trading down 8% the next day is shows that the market's a little shaky.

这家丹佛的公司销售用于集中管理分析海量数据的软件。他们现在开始招聘高中生，觉得大学毫无意义，直接从中学生里选拔人才。

The Denver based firm sells software to centralize, manage and analyze large amounts of data. They're hiring high schoolers now. They're just like college is pointless. We'll just hire you straight out of high school.

这种情况罕见得值得关注，《金融时报》这篇关于中美AI竞赛现状的文章确实不错。作者给'竞赛'打了引号。首先，对于AI这种通用技术而言，长期优势往往取决于技术在社会中的普及深度。文章指出中国在这方面有优势，因为他们过去三十年一直在做全社会层面的转型，这方面确实经验丰富。

This is rare enough to be highlighted and actually good Financial Times article on the state of the AI race between The US and China. He put race in quotes. One, the fact that for a general purpose technology such as AI, long term advantage often comes down to how widely and deeply technology spread across society. And China has advantages here as the article notes since they're unarguably better at orchestrating society wide transformation as that's all they've been doing for the past three decades. They got some reps in.

我尚未看到但希望了解的数据点是AI的实际应用情况。显然他们通过DeepSeek在尖端模型创建上已经迎头赶上，但这些模型真正落地应用的比例有多少？

The the data point that I have not seen that I would like to see is is actual consumption of AI. Like, it's clear that they were able to catch up on creation of frontier model with DeepSeek, but, how much is actually getting implemented?

这极具讽刺意味，因为美国实际上推动了中国开发更高效的模型。而且开源策略最终被证明更有利于全球技术扩散。

This is extremely ironic as The US effectively pushed China to develop models that are more efficient Yeah. And open source, which turns out to be a better strategy for global diffusion.

如果你

If you're

是德国或巴西的一家公司，可以下载Quine或DeepSeek，在本地运行，根据你的市场进行微调，完全不用担忧API调用限额问题。

a company in say Germany or Brazil, can download Quine or DeepSeek, run it locally, fine tune it for your market and never worry about API rate limits.

但这是否意味着你们就赢了？

But does that mean that you win?

文章指出，中国公众对AI的乐观程度位居世界首位。

The article notes, the public in China is the most optimistic in the world about AI.

英伟达是市值5万亿美元的公司。如果他们能向中国销售，市场规模真的能达到10万亿美元吗？有那么夸张吗？中国的市场机会究竟有多大？

NVIDIA is a $5,000,000,000,000 company. Like like, if they could sell to China, how much like, it's not like it's a $10,000,000,000,000 company. Is it? Is it that big of a deal? Like, how how how big is the China opportunity?

全球第二大计算市场。我不知道。这取决于你对AGI的依赖程度，约翰。特斯拉在中国销售的时间足够长，长到足以让他们复制它。

Second largest computing market in the world. I don't know. Depends on how AGI pulled you are, John. Tesla was selling into China as long as well, as long as it took for them to copy it.

纽约市正在Mamdani和Cuomo之间做选择，Mamdani有95%的胜算，Cuomo只有5%。

New York City is choosing between Mamdani and Cuomo, and Mamdani is at 95% chance. Cuomo's at 5%.

90。

90.

你知道，外面很多人说聪明钱都押在Cuomo身上。也许他会赢。

You know, there's lot of people out there saying smart money's on Cuomo. Maybe he'll win.

我觉得那些不靠谱的钱都押在Quoala上了。这可能是真的。Canva、Wingstop、Shake Shack等一堆餐厅的月度表现。这些餐厅股票最近还在以35到100倍的远期市盈率交易，很多现在依然如此。

I think the degenerate money is on Quoala. That's probably true. Monthly performance from Canva, Wingstop, Shake Shack, a bunch of other, restaurants. Most of these restaurant stocks were trading at 35 to a 100 times forward PE just recently. Many still are.

这只是泡沫在消退。泡沫不可能无限期持续。

This is just a bubble deflating. Bubbles can't be sustained indefinitely.

Chipotle对年轻消费者的大赌注正在瓦解。不确定的美国经济正在影响年轻一代美国人。这家连锁店多年来一直培养年轻客户群，将其卷饼和碗宣传为健身爱好者和健康饮食者的高蛋白营养餐。你怎么看，Ben？

Chipotle's big bet on younger consumers is is unraveling. Uncertain US economy is catching up to younger Americans. Chain has spent years cultivating a younger clientele, pitching its burritos and bowls as nutritious protein packed meals for gym bros and healthy eaters. What do you think, Ben?

在我人生的某些阶段，我90%的热量摄入都来自Chipotle。而且每次长途驾驶时，知道途中能停下来吃一碗新鲜营养的墨西哥碗，真是令人期待的事。

I I used to get easily 90% of my calories from Chipotle at certain times of my life. And and anytime you were, like, on a long drive. Yes. Knowing that at some point in the drive you could pull over and have a nice, fresh, nutritious slot bowl. It was something to look forward to.

哦，那太棒了。

Oh, it was amazing.

现在品质下滑到如此极端的地步，我宁愿禁食也不愿在几乎所有情况下吃Chipotle。

Quality is degraded to such an extreme degree that it's no, I'd rather fast than eat Chipotle in almost any situation.

某种回顾

Sort of look

我怀念那些把Chipotle当作奢侈品的日子，当时觉得'好吧'。

back fondly on the days when Chipotle was felt and was a luxury when I was like, okay.

加很多肉和鳄梨酱。额外加料时感觉自己像个国王。

Have a lot meat and guacamole. Extra guac and extra meat was like, I am a king.

你当时也是那种吃冷冻鸡胸肉的人吗？

Were you were you at, like, a frozen chicken breast guy too?

是啊，当然。

Yeah. For sure.

因为我当时正盯着那一磅装的米袋和冷冻鸡胸肉，感觉就像，又来了。又见面了，老兄。

Because I'd just be looking at I'd be looking at the one pound bag of rice and the frozen chicken breast being like, again. We meet again, man.

又见面了。

We meet again.

收入增长是因为他们在开新店，但在现有门店里，你会发现

Revenue is growing because they're they're opening new stores, but at existing stores, you see

他们得走高端路线，鱼子酱碗。能给我来份和牛吗？五份和牛碗，上面再加双份白鲸鱼子酱。历史刚刚被改写。Lux家族企业Andoril科技公司刚刚成功试飞了美国历史上首架全自主战斗机。

They gotta go up market caviar bowls. Can I have Wagyu? A five Wagyu bowl with, double beluga on top. History was just made. Lux family company, Andoril Technology, has just successfully flown the first ever fully autonomous fighter jet ever in US history.

这架战机外形确实很疯狂。

It it is a crazy looking jet.

主流说法是这架飞机是人类制造的。这些公司都是从哪冒出来的？我们强势回归了。凯西·汉默一直在说，我们在美国搞太阳能效率不高。但这家公司却火力全开，一天就能铺设14兆瓦的太阳能板。

The mainstream narrative is that this was created by humans. Where are all these companies coming from? We're so back. Casey Hanmer's been saying, like, we're not making solar in America effectively. And here's this company that's just, cooking, setting out 14 megawatts of solar panels in a single day.

马丁·斯克雷拉预测英伟达市值将达到100万亿美元。即便是处理邮件的文职工作也需要人类思考，但像法律、医疗、金融这类高薪工作同样可能被AI取代。

Martin Scrella is forecasting that NVIDIA will get to 100,000,000,000,000 in market cap. Even desk email jobs require human thought, but higher priced jobs like legal, medical, finance can also be replaced by AI.

这对英伟达的空头来说简直是噩梦。没错。这真的

This is a nightmare post for NVIDIA bears. Yeah. This is really

这可不是编的。SpaceX星舰27亿美元的HLS合同包含对接适配器，而NASA洛克希德的猎户座飞船已耗资3160亿美元且仍在增加。研发十九年，对接系统还要额外花费25亿美元。

Can't make this stuff up. SpaceX Starship, 2,700,000,000.0 HLS contract cost includes docking adapter, NASA Lockheed Orion, 31,600,000,000 cost and counting. 19 in nineteen years in development, docking is an extra 2,500,000,000.0 extra.

对接系统还要再加两成半费用。我们很抱歉。

The dock's gonna be an extra 2 and a half. We're sorry.

搞这么多就为了做个AI记事本应用排名第八第七什么乱七八糟的。

All this to be an AI notetaker app number eight seven blah blah blah blah blah.

问题在于，Cluely擅长营销的那类人群真的需要AI记事本吗？我不认为他们能做出比Otter和Granola好十倍的产品。

The question here is, is there a market for an AI notetaker for the kind of people that Cluely is really good at marketing too? I don't think it's possible for them to make a 10x better product than Otter and granola.

大学生实际在用什么呢？根据我们听到的，如果你想...比如作业作弊，会用什么工具？具体是什么组合，乔迪？其实是ChatGPT Atlas。2020年成立的Benchmark基金在AEX指数上标注为A级。

What are college students actually using? Like, what like, from what we've heard, the stack, if you want to, like, cheat on your homework. What is the stack, Jordy? It's actually ChatGPT Atlas. Benchmark 2020 vintage fund marked at an AEX.

泰勒·霍奇带来了报道。他们一切安好。

Tyler Hodge has the story. They're doing just fine.

我认为埃夫·兰德尔应该做的是，新的GP

I think what, Ev Randall should do, new GP

嗯。

Mhmm.

在Benchmark那边，他应该

Over at Benchmark What should he

做什么？

do?

又是那个基金的10倍。

Is 10 x that fund again.

你需要风险投资吗？打电话给Benchmark的埃夫·兰德尔。

Do you need venture capital? Call Ev Randall at Benchmark.

+1 8018筹集了1.5亿美元，为健康零食的下一波浪潮提供动力。

+1 8018 raises a 150,000,000 to power the next wave of healthy snacking.

这太疯狂了。

This is crazy.

Trudis完成了1.5亿美元的融资轮。这是个巨大的蓝莓项目。由摩根大通资产管理公司与雷·达里奥加倍押注。

Trudis has closed a $150,000,000 funding round. It's a huge blueberry. By JPMorgan Asset Management with Ray Dalio doubling down.

尝尝这些吧。

Try some of these.

这家浆果巨头估值超过10亿美元。这家以巨型蓝莓和即拿即走零食杯闻名的零食公司，去年仅蓝莓销售额就创造了超过4亿美元的收入。他们得在这里加入AI叙事。这就是把一个简单的想法认真对待，而有时这个简单的想法就是大得离谱的蓝莓。

Valuing the berry powerhouse at over 1,000,000,000. The snacking company best known for its jumbo blueberries and new grab and go snack cups generated more than 400,000,000 in revenue last year with blueberry sales. They gotta get an AI narrative here. This take a simple idea and take it seriously, and sometimes the simple idea is obscenely large blueberries.

Will.I.Am登上新闻是因为他将在亚利桑那州立大学教授一门关于自主AI的新课程。听听Will.I.

Will. I. Am is in the news because he will be teaching a new class on agentic AI at Arizona State University. Hearing how will. I.

Am如何运用AI会很有趣。我永远被那些有手艺的人所娱乐。比如，你可能不喜欢他的音乐，也可能热爱他的音乐，但很明显你知道他是做音乐的。他有一套创作流程。

Am is using AI would be interesting. I am endlessly entertained by there's somebody who has a craft. Like, you might not like his music. You might love his music, but clearly, you know that he makes music. Like, he has a process.

他有一间工作室。

He has a studio.

威尔·艾姆曾是特斯拉的早期投资者，在埃隆之前还投资过OpenAI、Anthropic、Twitter、Pinterest、Dropbox、Hugging Face、Beats和Runway。AI生成的歌曲首次登上了公告牌电台排行榜。扎尼亚·莫奈由人类塔莉莎·琼斯创作，她使用了Suno。

Will. Am was an early investor in Tesla, pre Elon as OpenAI, Anthropic, Twitter, Pinterest, Dropbox, Hugging Face, Beats, and Runway. AI generated song has entered a billboard radio chart for the first time. Zania Monet, created by human, Talisha Jones, who used Suno.

Suno。

Suno.

几周前还签了一份300万美元的唱片合约。想象一下刚跟进这事，突然就拿到了300万的唱片合约。我倒想看看，一个AI音乐人不公开真实身份，成为音乐界的'中本聪'。

Also signed a $3,000,000 record deal a few weeks ago. Imagine just following it up and then all of a sudden you got the $3,000,000 record deal. I would like to see, an AI musician not share their real identity and Satoshi of music.

音乐。不是流行巨星，而是'泥浆明星'。如果你想下线，想接触大自然，就去Wander吧，找到你的快乐角落，预订一个风景绝佳的Wander，体验升级菜单、梦幻床品、顶级清洁服务、24小时礼宾服务，度假或更好的

Music. Not a pop star, but a slop star. If you wanna log off, if you wanna touch some grass, head over to wander, find your happy place, book a wander with inspiring views, touch upgraded menus, dreamy beds, top tier cleaning, twenty four seven concierge service, vacation, or a better

和你们大多数人一样，我进入加密圈不是为了发财。我相信的是500家不同中心化机构在500条不同区块链上发行500种稳定币的愿景。

Like most of you guys, I didn't get into crypto to get rich. I got into crypto because I believed in the vision of 500 different stablecoins issued by 500 different centralized institutions institutions on 500 different blockchains.

呃。我确实想知道稳定结果是什么。最终的平衡点会是怎样？是无休止的新币泛滥，还是行业整合？

Ugh. I do wonder what the stable result is. Like, what what is the final equilibrium? Is it just endless proliferation of new coins, or is it a consolidation?

如果你持有1830亿美元的Tether，750亿美元的USCC。接下来最大的是90亿美元的USDE，其次是DAI，再往下是世界自由金融。

If you have Tether at a 183,000,000,000, USCC at 75,000,000,000. The next largest is USDE at 9,000,000,000 below that is DAI. And then below that, below that is World Liberty Financial.

我有一个初步形成的假设：艺术家对生成式AI的强烈反对，很大程度上是对文本驱动用户界面的反应。艺术界已经接受了像Photoshop这样的工具，以及更早的摄影本身，它们仍与传统艺术创作共享美学。如果Midjourney、ChatGPT、DALL·E等产品的界面能像2020年NVIDIA Canvas那样（我记得这个工具，它很棒），情况可能会不同。

I have this partially formed hypothesis that artist backlash against generative AI crystallized largely in response to text driven user interfaces. The art world has come to accept tools like Photoshop and earlier photography itself that still shares the aesthetic of traditional art creation. If mid journey, ChadGBT, Dolly, etcetera, had interfaces along the lines of NVIDIA Canvas from 2020 pictured below. I remember this tool. It was awesome.

你可以那样操作。随心所欲地绘制任何轮廓，它会自动生成并看起来很棒。Palmer认为，这样的界面足够接近现有工具，批评会显得纯粹而精英化。而通过命令行创作艺术，则显得格格不入，容易让人贬低使用者不是真正的艺术家。

You could do that. You could draw whatever whatever, like Right. Outline you wanted, and it would just automatically generate it and look great. Palmer says it would look enough like existing tools that that the critique would look purest and elite. Art via command line, on the other hand, looks sufficiently other that you can easily berate the users as not artists.

Service Titan市值90亿美元，营收8.66亿美元。与之相比，Harvey估值80亿美元，年经常性收入约1亿美元。至此，你必须相信我们正从垂直细分SaaS领域迈向下一场工业革命。你无法用现有法律AI市场的标准来评估Harvey——这本质上是对劳动力替代的赌注。

Service Titan at $9,000,000,000 market cap on 866,000,000 of revenue. Comparing that to Harvey, which is $8,000,000,000 valuation on somewhere around a 100,000,000 of ARR. At this point, you have to believe that we're moving from a world of niche vertical vertical SaaS to the next industrial revolution. You would not be able to underwrite Harvey like this against like the existing legal AI market. This is a bet on, labor displacement.

显然，全球200家最大律所每年提供1300亿美元的法律服务（前100强数据）。所以Harvey基本上已经触及这个规模。

Apparently, the 200 largest law firms in the world annually provide a 130,000,000,000 of legal services. That's the top 100. So Harvey's already sitting at, you know, basically there.

我们将

We will be

回到超能洛杉矶

back in the ultra gun Los

安吉利斯。我们待会儿见

Angeles. We'll see you

等不及了。那么。再见。

Can't wait. Then. Goodbye.