为什么我们今年夏天退出了英伟达，以及AI支出将流向何处 AI价值链将发生哪些变化？
Why We Exited Nvidia This Summer and Where AI Spending Goes Next What changes across the AI value chain?
AI相关股票的抛售在中期选举后加剧，原因是人们担心数据中心建设放缓。随着10年期美国国债收益率接近4.9%，借贷成本依然居高不下，投资者继续质疑AI支出的规模。
The sell-off in AI-linked stocks deepened after the midterms amid concerns about a slower data center buildout. With 10-year Treasury yields near 4.9%, borrowing costs remain high, and investors continue to question the scale of AI spending.
与此同时，公司报告称从AI中获得了强劲增长和回报。我们认为，这一时期将帮助区分胜出者和商业模式存疑的公司。
At the same time, companies are reporting strong growth and returns from AI. We believe this period will help separate the winners from companies with questionable business models.
- 为什么我们不持有英伟达本周最大的新闻是，在一波争议之后，黄仁勋辞去了英伟达的职务。英伟达历来被视为AI的代名词，因此他的离职引发了诸多疑问。
- Why We Don’t Own Nvidia The biggest news this week was Jensen Huang’s resignation from Nvidia following a wave of controversy. Nvidia has historically been viewed as a proxy for AI, so his departure raised many questions.
英伟达已任命首席财务官科莱特·克雷斯为临时首席执行官，同时寻找永久继任者。
Nvidia has appointed CFO Colette Kress as interim CEO while it searches for a permanent successor.
离职非常突然。我们认为，相关情况使得黄仁勋不太可能以其他身份继续参与。
The departure was abrupt. In our view, the circumstances make it less likely that Huang will remain involved in another capacity.
这个故事有很多角度（政治、伦理等），但我们认为，与我们的投资论点最相关的是出口管制。
There are many angles to this story (political, ethical, and more), but we believe the most relevant to our investment thesis concerns export controls.
背景是，2022年10月7日，美国宣布对向中国出口先进计算芯片实施广泛限制。此后，有指控称受限芯片通过第三国流向了中国。
For background, on October 7, 2022, the U.S. announced sweeping restrictions on exports of advanced computing chips to China. Since then, there have been allegations that restricted chips have been routed to China through third countries.
首席执行官通常与当地销售活动相隔数步。SMCI也面临类似指控，涉及流入俄罗斯和中国的芯片，其首席执行官受到的影响有限。然而，在黄仁勋的案例中，据报道他与一名前美加速公司高管有个人关系，我们认为这增加了另一层风险。这些指控仍在调查中。
CEOs are generally several steps removed from local sales activities. SMCI faced similar allegations related to chips that found their way into Russia and China, with limited repercussions for its CEO. In Huang’s case, however, his reported personal relationship with a former Megaspeed executive adds another layer of risk, in our view. The allegations remain under investigation.
黄仁勋的离职对英伟达是一大损失，我们认为，离职的时机使情况更加糟糕。
Huang’s departure is a loss for Nvidia, and we believe the timing makes it worse.
来自Anthropic和OpenAI的封闭模式正在赢得资金。
Closed models from Anthropic and OpenAI are winning the dollars.
我们认为，定制芯片将随着时间的推移占据更多这部分支出，从而限制英伟达的市场。虽然Anthropic和OpenAI目前都是英伟达的重要客户，但它们的算力承诺涵盖多家供应商，包括自行研发定制芯片。我们预计，这一动态将随着时间的推移对英伟达越来越不利。开源模式正在赢得规模。
Our bet is that custom silicon will capture more of that spending over time, limiting Nvidia’s market While both Anthropic and OpenAI are significant Nvidia customers today, their compute commitments span multiple providers, including building their custom chips. We expect this dynamic to become less favorable for Nvidia over time. Open models are winning the volume.
这通常有利于英伟达的通用GPU，但中国模式正在占据显著市场份额。进一步的出口限制可能会将英伟达挤出这个市场。
That would typically be good for Nvidia’s general-purpose GPUs, but Chinese models are taking significant market share. Further export restrictions could shut Nvidia out of this market.
封闭模式公司也从较少的中国竞争中受益，并且是出口限制的主要推动者之一。
Closed model companies also benefit from less Chinese competition and are one of the main proponents of export restrictions.
总结来说，历史上，英伟达占据80%的市场份额，是AI芯片领域无可争议的领导者。随着封闭模式在今年初开始主导AI支出，动态发生了变化。更多的AI并不自动意味着更多的英伟达。
In summary, historically, Nvidia held an 80% market share and was the undisputed leader in AI chips. As closed models began to dominate AI spending at the start of this year, the dynamics changed. More AI does not automatically mean more Nvidia.
- Anthropic即将进行的IPO Anthropic周二提交了其IPO条款：每股70至78美元，完全稀释后估值约为1.26万亿至1.41万亿美元。这远低于今年夏天传闻的约2万亿美元，也略低于我们此前预期的1.5万亿美元，当时AI估值正在重新定调。价格下降了，但业务没有：估值：按74美元的中点计算，Anthropic的估值将比其5月份的融资轮次高出38%（当时为9650亿美元）。
- Anthropic’s Upcoming IPO Anthropic filed its IPO terms on Tuesday: $70 to $78 a share, about $1.26–1.41 trillion fully diluted. That’s well below the ~$2 trillion talked about this summer, and just under the $1.5 trillion we had expected as AI valuations reset. The price came down. The business didn’t: Valuation: at the $74 midpoint, Anthropic would be worth 38% more than at its May round ($965 billion).
收入：季度收入在同期内增长了3.6倍，达到第三季度的170亿美元，且截至10月底，年化收入约为810亿美元。
Revenue: quarterly revenue grew 3.6-fold over the same stretch, to $17.0 billion in Q3, and run-rate revenue reached about $81 billion at the end of October.
倍数：5月份的投资者支付了约为季度年化收入51倍的价格。新的区间相当于约20倍。定价预计将于11月18日确定，交易将从19日开始。
Multiple: May’s investors paid about 51 times annualized quarterly revenue. The new range works out to about 20 times. Pricing is expected on November 18, with trading from the 19th.
背景信息：“完全稀释”是指计算所有可能存在的股份，包括期权和未结算的股票奖励。仅计算发行后的流通在外股份，同样的74美元价格意味着约9680亿美元的估值。
For background: "fully diluted" counts every share that could exist, including options and unsettled stock awards. Counting only the shares outstanding after the offering, the same $74 price implies about $968 billion.
最引人注目的是还有谁在购买。在本次发行之际，谷歌和英伟达（各100亿美元）、AMD、博通、美光、微软和台积电将直接购买260亿美元的股票：涵盖云服务、GPU、定制芯片、存储和晶圆代工。他们正在投资一家拥有约9010亿美元承诺支出的客户，其中几乎全部用于计算。
What stood out to us is who else is buying. Alongside the offering, Google and NVIDIA ($10 billion each), AMD, Broadcom, Micron, Microsoft and TSMC will buy $26 billion of stock directly: the cloud, the GPU, the custom chip, the memory and the foundry. They’re investing in a customer with about $901 billion of committed spending, nearly all of it on compute.
OpenAI也在筹资：上周，它完成了110亿美元的融资，估值达1.25万亿美元，较3月的8520亿美元有所上升，这是一轮可能达到约200亿美元融资的第一部分。
OpenAI is raising too: last week it closed $11 billion at a $1.25 trillion valuation, up from $852 billion in March, as the first part of a round that could reach about $20 billion.
我们的看法是：IPO为最大的新计算采购者提供了公开货币和新资本，其中大部分将流向计算能力、芯片、存储和网络等下游环节。
Our read: the IPO gives the largest buyer of new compute a public currency and fresh capital, most of which flows down the stack into compute capacity, chips, memory, and networking.
批评者称供应商投资是循环融资。我们认为，供应商正在确保其未来最大的订单来源。
Critics call supplier investments circular. We see suppliers securing their biggest source of future orders.
- 价值链上会发生哪些变化？黄仁勋辞职以及英伟达竞争地位可能减弱，对人工智能价值链有着重要影响。
- What Changes Across the Value Chain? Jensen’s resignation and a potentially weaker competitive position for Nvidia have meaningful implications for the AI value chain.
我们不认为整体人工智能支出会受到影响，考虑到生态系统已经相当成熟。但我们认为，一些公司将受益，而另一些公司将面临更大压力。
We do not expect overall AI spending to be affected, given how developed the ecosystem has become. But we believe some companies will benefit while others face greater pressure.
循环融资将接受现实检验。任何指望英伟达救助的人都应重新审视自己的假设。
Circular financing gets a reality check. Anyone counting on an Nvidia bailout should revisit their assumptions.
英伟达支持了一个既包括供应商也包括客户的生态系统。客户，特别是依赖英伟达支持的新云服务商，可能受影响最大。
Nvidia has supported an ecosystem of both suppliers and customers. Customers, particularly neoclouds that rely on Nvidia’s backing, could be the most affected.
我们对英伟达在模型和应用层的投资不太担心。我们认为，这些交易主要是由公司确保计算供应的需求驱动，而非融资需求。竞争对手和供应商可能从更多元化的人工智能芯片组合中受益。AMD和博通等定制芯片供应商是天然的受益者。
We are less concerned about Nvidia’s investments in the model and application layers. In our view, those deals were driven primarily by the companies’ need to secure compute supply, rather than financing. Competitors and suppliers could benefit from a more diversified mix of AI chips. AMD and custom-chip providers such as Broadcom are natural beneficiaries.
网络是另一个我们认为市场低估的潜在受益方。英伟达构建了一个与其服务器系统紧密集成的强大网络平台。AI芯片组合更加多元化可能削弱这种产品锁定效应，提升独立网络供应商的需求，并使其拥有更大的定价权。
Networking is another potential beneficiary that we believe the market underappreciates. Nvidia has built a strong networking platform that is tightly integrated with its server systems. A more diversified mix of AI chips could reduce that product lock-in, boosting demand for independent networking suppliers and giving them greater pricing power.
底线 我们对AI支出持乐观态度，但支出流向以及谁能获得回报正在发生变化。更多的AI投资并不自动意味着更多的英伟达，我们认为在计算和网络领域都存在机会，因为客户正在多元化其供应链。我们关注的公司是那些能够将这项投资转化为持续增长和回报的公司。
Bottom Line We remain constructive on AI spending, but where that spending goes, and who captures the returns, is changing. More AI does not automatically mean more Nvidia, and we see opportunities across compute and networking as customers diversify their supply. Our focus is on companies that can turn that investment into durable growth and returns.
我们正在关注 英伟达11月17日的业绩。Anthropic的定价预计将于11月18日公布，其首次交易将于19日进行。以及美联储12月会议，市场预计加息的可能性约为80%。
What We’re Watching NVIDIA’s results on November 17. Anthropic’s pricing, expected November 18, and its first trade on the 19th. And the Fed’s December meeting, where markets see about an 80% chance of a rate hike.