你早就知道这事得发生。图表早已注定。我说的是人工智能交易各个方面的崩盘。你能感觉到它在 Meta 的图表上悄悄逼近。接着出现了市场无法承受的那一件事：巨额潜在融资，而我们原以为支持 Muse 的基础设施建设将是自筹资金的。并非如此。所以一波巨大而漂亮的涨势已基本消失。微软和 Alphabet 周三、亚马逊周四会不会遭遇同样的命运？
You knew it had to happen. The charts foreordained it. I am talking about the collapse in every aspect of the artificial intelligence trade. You could feel it creeping up on the chart of Meta . Then we got the the one thing the market couldn't bear: a huge prospective capital raise, when we thought the buildout of infrastructure to support Muse was to be self-funded. Hardly. So a gigantic, beautiful move has mostly disappeared. Could the same be coming for Microsoft and Alphabet Wednesday, or Amazon on Thursday?
也许？也许不会，但上行空间永远不会超过下行风险，因为戏已经唱完了。给我金佰利、可口可乐，当然还有宝洁。它们现在能把那堆脆弱的硬盘驱动器和重型资本设备甩在身后。我们真的还在乎 Lam Research 吗？Applied Materials 的危险只能被 Broadcom 或 Seagate 超越。我从未吸过海洛因，谢天谢地。我有另一种瘾：对这组一生难遇的惊人表现的令人麻木的忠诚。
Maybe? Maybe not, but the upside would never trump the downside because the jig was up. Give me Kimberly-Clark or Coca-Cola and most certainly Procter & Gamble . They can now beat the pants off the complex of flimsy disc drives and heavy capital equipment. Did we really care about Lam Research anymore? The danger of Applied Materials can only be outdone by Broadcom or Seagate . I've never tried heroin, thank heavens. I had a different addiction: a mind-numbing fealty to the incredible performance of this once-in-a-lifetime group.
为什么不呢？过去几个月，AI 似乎在无论什么领域都比我们更聪明。我们只是不知道罢了，但我们会知道的。它正在取代大量白领工作，同时也为急需建设数据中心的蓝领工人创造了许多就业岗位。但问题首先出在数据中心上。城市和州过去争相招引它们。就业稀缺的城镇欢迎它们。第一批数据中心引发了某些地区对水源污染和电价上涨的担忧。
Why not? The AI had seemingly become smarter than us these last few months, no matter what the field. We all just didn't know it, but we will. It is taking the place of lots of white-collar people, even as it creates many jobs for the blue-collar folk who are urgently needed to build the data centers. But there's the first problem: the data centers. Cities and states used to bid for them. Towns with few jobs welcomed them. The first tranche caused concerns in certain locales about despoiled water and increased electricity rates.
等到数据中心建设者意识到必须付钱给城市以压低价格、必须给包括教师在内的城市工人发奖金时，似乎已经太晚了，至少赶不上选举了。妖魔化随之开始且从未停止。它们似乎比移民与海关执法局（ICE）还不受欢迎。也许它们能熬过选举，也许不能。
By the time the data-center builders figured out that they had to pay the cities to keep prices down, that they had to bonus city workers, including teachers, it almost seemed too late, at least for the elections. The demonizing began and never ended. They seemingly became less popular than Immigration and Customs Enforcement. Maybe they survive the elections, maybe they don't.
他们拥有了这位有条不紊的冠军——英伟达的杰出掌舵人黄仁勋，以及他公司令人惊叹的 AI 芯片，即图形处理单元（GPU）。人人皆知这些芯片最出色，但只有埃隆·马斯克毫无保留地大量采购，并承诺专用英伟达芯片。其他超大规模云服务商想要这些芯片，是因为客户指名要用，但它们都大肆吹嘘自家的定制芯片。我最近看到有说法称亚马逊的 Trainium 在某些基准测试中优于英伟达。
They have their methodical champion, the great Jensen Huang of Nvidia, and his company's amazing AI chips, known as graphics processing units (GPUs). Everyone knew they were the best, but only Elon Musk had no reservations buying them, and he's pledged to exclusively use Nvidia chips. The other hyperscalers wanted them for the clients that insisted to have them, but they all crowed about their own custom chips. I saw something recently about Amazon's Trainium being superior to Nvidia in certain benchmarks.
我不信这个说法，但那又怎样？这引出了故事背后的第二支“暗箭”：打造一个生态系统，让我国在生成式 AI 和加速计算领域成为最强国家。我们必须击败中国。我们需要击败中国。政府会出手吗？不会，除了挽救陷入困境的英特尔，不会有任何政府支持。一切将靠英伟达，它欣然承担起这个角色，为数据中心的方方面面提供资金，包括亲自盖那些“庞然大物”。
I don't believe the claim, but what does that matter? Which brings me to the second arrow in the back of the story: the creation of an ecosystem that would make our country the superior country when it came to generative AI and accelerated computing. We needed to beat China. We need to beat China. Was the government going to do it? Nope, with the exception of a save of the struggling Intel , there would be no government support. It would be up to Nvidia, which embraced the role, bankrolling all sorts of data center aspects including the actual building of the beasts.
于是英伟达资助一些公司，被称为“新云厂商”，这些公司转手用这笔钱采购英伟达产品。对某些人来说，这一过程被视为循环融资。英伟达通过股权投资和合作伙伴关系提供支持，随后 CoreWeave、Nebius 和 Iren 等公司采购最新最强的英伟达芯片——现役的 Vera Rubin 之前是 Grace Blackwell、Hopper 和 Ampere。
So Nvidia funded companies, called neoclouds, which would turn around and buy Nvidia product with the money. For some this process was deemed circular. Nvidia offers support via equity stakes and partnerships, then CoreWeave , Nebius and Iren , among others, buy the latest and greatest Nvidia chips — the current generation, Vera Rubin, was preceded by Grace Blackwell, Hopper and Ampere.
服务器制造商超微电脑、戴尔科技和惠普企业抢购一空，CoreWeave 和其他新云厂商也是如此。我们知道戴尔、超微和 HPE 有支付能力。其他公司没有。它们会接受供应商融资。怀疑者视这部分革命为薄弱环节。这似乎从未成问题，直到 CoreWeave 股价开始像 K2 峰一样令人胆寒地暴跌，其高收益债务成为祸首。
Instantly sold out as server makers Super Micro Computer , Dell Technologies and HP Enterprise took whatever they could get. So did CoreWeave and the other neoclouds. We knew that Dell, Super Micro and HPE could pay. Not the others. They would take the vendor financing. The skeptic viewed this part of the revolution as a weak link. It never seemed to matter until CoreWeave's stock began a sickening K2-like fall, with its high-yield debt being the culprit.
CoreWeave 是一位关键客户，即使其英伟达芯片的使用寿命超过了折旧期限，其大部分价值似乎也已被预先占定。这个资金不足的实体竟能引发系统性崩盘，似乎有点不可思议。但如果没有 Anthropic 和 OpenAI 即将进行的首次公开募股（IPO）带来新资金，让这些 AI 实验室能继续支付自身账单，人们开始担心 CoreWeave 偿还债务的能力。
CoreWeave was a vital client, and even as its Nvidia chips lasted longer than its depreciation , much of the value seemed spoken for. It seemed a little weird that this undercapitalized entity could bring the house down. But without fresh money coming from the prospective initial public offerings of Anthropic and OpenAI to allow those AI labs to keep paying their own bills, people started to worry about CoreWeave's ability to service its debt.
它们的交易似乎越来越遥不可及。最终这甚至都不重要了。空头 1 分，多头 0 分。现在我们知道 AI 不会消失，远非如此。但由于几乎没人能为其买单——因为除了英伟达，它几乎不为任何人赚钱——股市嗅到了血腥味，想要榨干这一产业链的资本。这是铁路的第一次崩盘。它们将重新启动，只管等着瞧吧。但没人在等了。
Their deals seem to grow further and further away. In the end it didn't even matter. Bears, one. Bulls, zero. Now we know AI isn't going away. Far from it. But with so few to pay for it, because it makes so little money for all but Nvidia, the stock market smells blood and wants the complex robbed of its capital. It's the first collapse of the railroads. They will be started again, just you wait. But nobody's waiting.
谁还想做那个为这些数据中心项目输血的最后冤大头？这是一个卑鄙的词汇，却涵盖了经济增长和股市的很大一部分份额。它的触角伸得如此之远，以至于你不得不买药股和肥皂股才能彻底避开它。现在这就像一场非常严重的流感，你只能通过抛售一切长期以来行之有效的资产来给自己接种疫苗。这还会再次奏效吗？
Who wants to be the last one still funding these data center projects, a dastardly term that encompasses a significant share of economic growth and the stock market? Its reach is so vast that you have to go to drug and soap stocks to stay free of it. It's now like a really bad flu, one that you can immunize yourself from by jettisoning everything that worked for so long. Will it work again?
如果你留下来想弄个究竟，就会失去免受批评的护身符。谁想招惹这种事？我们想试着给这板块做做心肺复苏，但图表说别费劲了。图表会错，只是它们不仅仅是由虚无缥缈的 CoreWeave 推动的。它们是由更高的利率、更高的油价和一个反复无常的总统推动的，后者通过控制国会和军队将两者都推高。无论你恨他还是喜欢他，你都得问一句：他到底在干什么。
You lose your immunity from criticism if you hang around to find out. Who wants that to happen? We want to try to figure out how to give the group CPR, but the charts say don't bother. The charts would be wrong, except they aren't just fueled by the gossamer CoreWeaves. They are fueled by higher rates and higher oil and a fickle president that drives them both up by controlling Congress and the military. Hate him or like him, you have to ask what the heck is he doing.
对我们而言，我们竭尽全力想减少对 AI 的敞口，尽管这意味着要放弃巨额资本利得，因为根据我们发誓遵守的慈善信托指定，这些收益必须捐出去。
For us, we have tried mightily to trim our exposure to AI, despite taking huge capital gains that must be given away, as per the charitable trust designation we have sworn to.
我们不是对冲基金——干过那行，懂那套路——而我们手里剩下的几乎所有持仓都坐拥巨额收益，只有英特尔和存储芯片巨头美光除外。美光赚得盆满钵满，迎接它的却是空头们“见顶了、见顶了、见顶了”的一片叫嚣。我是不是该把这声音加进《疯钱》节目的音效板里？赌注大得惊人。我们得熬过大选。还得熬过总统会不会、怎么对付伊朗。
We aren't a hedge fund — been there, done that — and we have gigantic gains on almost everything left, save Intel and memory-chip giant Micron . Micron is earning huge amounts of money, only be greeted with the sound of bears claiming peak, peak, peak. Maybe I should add that to the soundboard on "Mad Money" ? The stakes are so high. We have to get past the election. We have to get past what the president will or won't do with Iran.
数据中心的鼓鼓囊囊的订单量，可能会变成数据中心的船底舱——烂泥一堆。我们已经摸透了美联储主席凯文·沃什的底细，他清楚这个国家打工人的微薄积蓄必须得到保全，他将通过抗击通胀来为他们守住这笔钱。如果油价腰斩，他就有回旋余地暂缓进一步加息。或许CoreWeave之流能融到资。或许订单持续涌向英伟达。
They may be too hard to ride through with a bulge of data center turned into the bilge of data center. We have taken the measure of Federal Reserve Chairman Kevin Warsh, and he knows that the working person's meager savings in this country must be preserved and he's going to preserve it for them by fighting inflation. If oil gets cut in half, then he has leeway to hold off on more rate hikes. Maybe the CoreWeaves of the world can raise money. Maybe the orders keep flowing to Nvidia.
这一整段时期里，我们看到只有一只股票持续上涨：苹果。空头们不断堆砌零部件调查或供应链“渠道检查”当借口，而这些几乎总是过度悲观。苹果设备阵容或许是史上最强，配得上我深深敬佩的一个人留下的惊人遗产：蒂姆·库克，他在9月1日把CEO火炬传给了约翰·特努斯。iPhone Duo可能很贵，但它是终极开户神器。
Throughout this period we have seen one stock continue to rally: Apple . The bears keep piling on about component surveys or supply chain "channel checks," which have almost always been too negative. The lineup of Apple devices might be the strongest ever, befitting the amazing legacy of a person I so admire: Tim Cook, who passed the CEO torch to John Ternus on Sept. 1. The iPhone Duo may be expensive, but it is the ultimate account opener.
随着SpaceX疯狂构建无线网络，某家电话运营商——大概率是T-Mobile——将吞下利润空间，碾压其他公司，赢得这场战役。SpaceX能不能赢下整场战争谁也说不准，但凭借马斯克购买的海量频谱和他的融资能力，Starlink极有可能先从Charter和Comcast手里抢份额，再从AT&T和Verizon手里抢份额。要说有条银线，那就是10月初以来家得宝的股价。一家专注住房的零售商本该下跌，不该上涨。它现在是一座灯塔。我就拿它了。
And with SpaceX on a rampage to build a wireless network, some phone company — presumably T-Mobile — will eat the margin and crush the other companies and win the battle. Who knows if SpaceX will win the war, but with the amount of spectrum being bought by Musk and his ability to access capital, Starlink will most likely take share from Charter and Comcast first, and then AT & T and Verizon next. If there is a silver lining, it's the stock of Home Depot since early October. A retailer devoted to housing should be going down, not up. It's a beacon right now. I'll take it.
我并不怀疑本周出现反弹的可能性。微软比Meta更有纪律、更对股东友好。亚马逊拥有一个假设性的500亿美元芯片业务，并能提供通往巨额盈利的路线图。Alphabet可以阐述为何需要如此庞大的算力，以及能从中赚取多少利润。这些因素中的任何一个都能引发一波反弹。这些因素中的任何一个都能引发一波反弹，但我预期这些反弹会昙花一现，因为我们并未超卖，且池子里的人依然太多。
I am not doubting the possibility of a bounce this week. Microsoft is more disciplined and shareholder friendly than Meta. Amazon has a hypothetical $50 billion chip business and can offer a roadmap to immense profitability. Alphabet can talk about why it needs so much computing power and how much it can make for it. Any one of these can spark a rally. Any one of these can spark a rally, but I expect those rallies to fade because we aren't oversold and too many are still in the pool.
我们已竭力精简持仓，并抛售了每一只抛物线式上涨的股票，只留下苹果和英伟达。我们将继续削减那些上涨的仓位，并聚拢马车“优化”投资组合。为何不抛弃一切？因为这场革命是真实的。我们无法抛弃织布机、铁路或个人电脑。那些革命最终造就了赢家。但在那之前，它们制造了无数输家。我们倾向于选择建设发电厂的安全性，通用电气Vernova（GE Vernova）便参与其中。
We've done our best to trim and we have sold every parabolic mover, save Apple and Nvidia. We will keep trimming the ones that go higher and circling the wagons to "high grade" the portfolio. Why not abandon everything? Because the revolution is real. We can't abandon the loom or the railroad or the personal computer. Those revolutions produced winners. But not until they produced losers galore. We like to be with the safety of building power plants, where GE Vernova is involved.
我们知道半导体将永远需要Qnity的材料。高带宽存储和CPU将供不应求。但持有美光和英特尔只会在有益之前先造成损害。亏损的幽谷太深，大多数人难以承受。所以我们竭力持有部分曾让进场又离场者大富的标的，尽管那只是寥寥无几的幸运儿。我们认为，AI是真实的，但除英伟达、AMD、戴尔和HPE外，若资金枯竭、OpenAI和Anthropic不来救援，其他公司的利润皆是虚幻且不可持续的。
We know the semiconductors will always need the materials of Qnity. High-bandwidth memory and CPUs will be in intense demand. But owning Micron and Intel can only hurt before they can help. The valley of the shadow of losses is too deep for most. So we do our best owning some of what made everyone who got in and got out so rich, although that's a fickle few. And we say, AI is for real, but the profits for all but Nvidia, AMD, Dell and HPE are chimerical and unsustainable if the capital runs out and Open AI and Anthropic don't come to the rescue.
眼下形势颇似卡斯特战役。或许这一次骑兵能及时赶到。但还能剩下多少人？我们只能拭目以待。
It seems Custer-like right now. Maybe in this version the cavalry gets there on time. But how many will be left? We'll just have to see. ()