Bloomberg Big Take播客复盘CoreWeave上周因举债担忧引发的股价/债券抛售，无新增可交易事实

【播客逐字稿全集 · Bloomberg Big Take】
单集:Musk Is Spending Big to Help Republicans — Here’s What He Wants
说明:以下为该集逐字稿的**完整原文**(未删改;英文档为 YouTube 自动字幕/节目原文,可能有识别误差)。请就本集内容判断是否构成**可交易**事件 —— 公司/产品/订单/产能/需求/合作/资本开支/高管表态等;泛泛的观点、预测、行业感想不算。若本集无可交易内容,判为不做。

--- 全集原文 ---
Speaker 1: Bloomberg Audio Studios.
Speaker 2: Podcasts.
Speaker 1: Radio.
Speaker 2: News.
Speaker 3: There's a lot of factors that are creating this perfect storm for why the market was so reactive. Investors are already showing a lot of fatigue around these data center deals and the amount of capital that they require. And, you know, we're seeing a lot of that investor FOMO be replaced with skepticism.
Speaker 1: It's possible that by next week, this will have blown over. It'll be fine. Corey will be stable and some other company will be raising questions or nobody will be. I think what's clear in any of these sorts of boom periods is that there's no situation in which all these companies make it.
Speaker 2: I'm Sarah Holder, and this is The Big Take from Bloomberg News. Today on the show, I sit down with Bloomberg's Catherine Schwartz, who covers credit, and Dina Bass, who covers AI infrastructure, to talk about the CoreWeave sell-off, what it means for neoclouds, and the future of the AI build-out. So the big story in AI right now is what's going on with a company called CoreWeave. It saw its stocks plummet as much as 17% last Wednesday. By the end of the week, it was down a total of 23.5% since the beginning of last week. Catherine, what happened?
Speaker 3: One main catalyst is that the company's co-founder said that CoreWeave was exploring ways to borrow in investment-grade markets and explore those pools of capital, which is basically creating a lot of lender concerns that the company is planning to borrow more at a time when there are already concerns about its leverage and debt capacity.
Speaker 1: CoreWeave is part and sort of was the original company in this category that people love to call neoclouds. CoreWeave hates that title. partly because they feel that they're not a new cloud anymore. They've been selling this kind of AI capacity for, you know, seven, eight years now at predating ChatGPT, to be clear. And so they feel like the new part is a little weird. They're also trying to kind of get out of people thinking of them purely as software somebody you can go to for GPU capacity. They're trying to sell a lot more software and cloud services on top of it. Things like that are more profitable. The reason we're talking about CoreWeave debt is because every time they want to get another data center up and running that they can sell capacity in, they end up having to borrow quite a bit to do that. It's pretty capital intensive.
Speaker 2: Dina, what is a neocloud? What role do neoclouds like CoreWeave serve in the AI space?
Speaker 1: They're really AI-focused clouds. So when you think back to the history of cloud computing and the companies that most of us think about, Amazon Web Services, Microsoft Azure, Google Cloud Platform, those are companies that started in an era that was pre-generative AI. They were selling cloud services to fuel all sorts of things, like your email, your accounting system. And more recently, they've been selling AI services. But they do cloud computing for all sorts of different workloads and tasks. The Neo clouds, for the most part, are AI clouds. They're focused around providing computing on top of NVIDIA GPUs, which are chips that get used for AI. Really, this whole industry grew up around how do you get customers enough NVIDIA GPUs that they can train and then run their models.
Speaker 2: And Catherine was describing the sort of perfect storm for CoreWeave, why its stock was plummeting, why it's moving into distressed territory. But how are other companies in the category faring right now? How does CoreWeave stack up?
Speaker 3: Yeah, I would say kind of building off of what Dina said, there is a lot of fear, first of all, tied into these neoclouds because for one reason, they also tend to be younger companies with higher yield ratings. So some of the losses on CoreWeave's debt, you know, kind of undermine some of the appetite for the neocloud model itself. And something that we're hearing from investors is for a neocloud deal to clear the market right now, it's going to be really difficult or even prohibitively expensive. So for one, there's a lot of concentration of fears around neocloud, but there's so much issuance or oversupply in the market, and that's outweighing investor demand. And the result of that is basically, one, you have so much kind of investor fatigue now around some of these data center financings. So I think that's where a lot of the tension is residing right now.
Speaker 1: Right.
Speaker 2: So it's not just a Coral Reef problem. It's not just a Neocloud problem. It's a problem that a lot of these AI companies are building this infrastructure on corporate debt and the corporate debt market is saying, oh.
Speaker 3: Yeah, we're seeing a lot of that kind of where there was a lot of investor FOMO around AI and everyone wanted to get involved and it was this hot thing. We're seeing a lot of that be sort of replaced with skepticism now.
Speaker 1: Can I differentiate on that last question between two different sets of demand here that I think is important? I think there's the debt picture that Catherine's talking about, and then there's the demand for AI compute, which is the flip side of it, but is not seeing a slowdown. So one of the things that we're seeing here that's very interesting is that CoreWeave is largely sold out of capacity. Most of the neoclouds that report publicly are saying the same thing. So we're ending up in a position where The demand for AI services for AI compute is still very high, and there's not a lot of excess in the system. But the issue is, how are you going to fund all of that? Part of the challenge is that you have both data center construction, which is already quite expensive, and then the cost to purchase the chips to put in that data center ends up being a couple of times, in many cases, the cost of the data center debt.
Speaker 2: That's an issue for the industry writ large, right, Dina? That's not just a core weave problem. That's sort of a headwind for the neocloud business model.
Speaker 1: For sure. And look, I mean, there's a big change here, and that's part of what people have been grappling with. You know, again, we were talking before about the early days of cloud. In the early days of cloud and even the early, you know, the early couple of years of the AI buildout, when you were talking about a Microsoft or an Amazon or a Google, they were– able to pay for these things off of their own balance sheet, either through their cash or they could take out debt on their own investment grade ratings. They were fine. When you end up with not just neoclouds, but also OpenAI and Anthropic, they don't have the ability to borrow for these things. OpenAI is starting to make their own AI chips and they have to figure out how to pay for that. And so what ended up happening is we moved to a model of, you know, sort of more like a project financing model. The folks at Cori, the co-founders, will tell you, look, That's how capital projects always get built. Most of the co-founders, of course, have come out of the commodities industry. And their argument is, look like when you're building an LNG plant or something like that, that's what you do. You sell the contract, and then you raise money on the back of that contract that you already have in order to build the facility. But even though it's been going on for a couple of years now, it's still a newish model for the technology space. And for the last couple of years, investors have seemed like they were willing to, you know, to buy into that. And now I think there's some questions.
Speaker 2: Katherine, you've been speaking to investors about CoreWeave's situation. And, you know, the biggest spike in yields really came after CoreWeave's co-founder said the company wanted to get access to, quote, unlimited pools of capital by getting this investment grade rating. Why was the suggestion that CoreWeave would be borrowing more such a red flag?
Speaker 3: Yeah, I think CoreWeave's debt has grown at a pace that is pretty extraordinary, even among its AI infrastructure peers, where you're already dealing with a lot of borrowing. I think compared to its peers, it's materially more levered in some ways, at least compared to some of the public data center comps. And it carries a pretty low interest coverage in the group. So I think that high leverage and then also a main distinction of core revisits dependent on outside financing, as Dino was sort of explaining. So unlike some of those heavy cash hyperscalers like Meta and Amazon, it doesn't have a diversified business that's funding the AI investment internally. So that's making it a lot more exposed to refinancing risk. And the thing about its tons of billions of debt is now also we're at a time of a higher interest rate environment, which is going to make the cost of borrowing more expensive. So I think that's sort of compounding a lot of those investor concerns.
Speaker 2: And Bloomberg did report on this internal email from Corweave CEO Michael Intrader, who said the company's backlog, quote, more than supports its debt obligations. But Catherine, again, you've been talking to investors who are saying this actually signals something sort of deeper about the corporate debt market more broadly. What is the concern?
Speaker 3: Yeah, a lot of the investors I speak to basically say that's kind of missing the point. This isn't just a Corweave problem. This is a contagion that is spreading to a lot of the data centers and the debt backing these data centers. And a lot of what I'm hearing is it's not just a credit profile problem, and it's not just that these data center deals might have changed in the last year since their issuance that investors signed on. It's just that we have another year and a half of supply coming. There's at least hundreds of billions of more issuance in the market expected to come to market in the next year, and the market's already drowning in this supply. So when you think about what is this worth now From an investor's perspective, you're thinking maybe I should wait on the sidelines if I know I'm going to lose money in the near term on this.
Speaker 2: We'll hear more about the contagion after the break. As CoreWeave stock was falling last week, Jensen Wang, the CEO of NVIDIA, came out and he tried to reassure people about the company's stability. Dina, why was that?
Speaker 1: Okay, one most basic thing here. Jensen Huang and NVIDIA are huge CoreWeave shareholders. They put $ 2 billion into CoreWeave earlier this year. It is one of their businesses. you know, significant customers. Many of these Neo clouds are very significant NVIDIA customers. There are a couple that are set up around the Google chip ecosystem and a couple set up around the AMD chip ecosystem, but the majority of the Neo clouds run entirely on NVIDIA GPU. And you have to realize that in order for them to continue to grow at the rate that they have forecast and at the rate that they have grown thus far, They need their chip customers to be able to pay for their chips. So that means CoreWeave has to be able to buy GPUs. And that's why you see Jensen out there defending CoreWeave.
Speaker 2: Big picture, Dina, are people stopping believing in the way that the AI build-out has been financed? You know, as Catherine alluded to, CoreWeave's CEO has said the market just misunderstands what's going on. Like, he's making this huge bet on the future. He needs to borrow this money to finance the build-out. And the potential on the other side is massive.
Speaker 1: Right.
Speaker 2: Does what's happening with CoreWeave mean investors are starting to question this bet in a new way?
Speaker 1: It seems like people are starting to pull back in some things. You know, over the weekend, Bloomberg reported that Nscale, another neocloud based out of the UK, which has filed to go public, that there's been pressure on Nscale from some of their investors to lower the range for their IPO. There was an IPO out of Australia from another NVIDIA-backed neocloud that was pulled. So you can see starting, you know, the beginning of concerns, at least about the pricing on some of the, you know, how some of these companies are valued. Does it last? It can be a little hard to say. I think we saw in July with the situational awareness, you know, blow up that there was this moment of people pulling back, you know, for at least a week or two. And then it kind of faded a little bit. So we'll have to see whether this continues in a more long term sense.
Speaker 3: One thing I would add as well to that list of the neocloud fears is last week, there was this deal for Volta Infrastructure, which is another AI startup, neocloud, and it priced this 5 billion leverage loan where the banks, JP Morgan and Goldman, had to really sweeten the terms to attract investors. I think it ended up being among the highest rates paid ever to finance the AI build out. So I think that's also another fresh barometer of, you know, the market's risk premium for the NeoCloud model.
Speaker 1: And Volt is an interesting case because they're a very young NeoCloud. They were only founded this year. They already raised a fairly rich valuation earlier this summer. Their whole pitch was, look, we're, you know, we're not just a NeoCloud. We will help our compute capacity customers get financing. But, you know, to Catherine's point, they're finding it, you know, more challenging themselves in terms of the rates they're having to pay to get financing.
Speaker 2: Well, At the end of last week, Dina, CoreWeave's bonds were dipping into distressed territory. What are the companies that are most exposed to CoreWeave right now? And if the CoreWeave sell-off were to continue or if things got worse, who might go down with them?
Speaker 3: NVIDIA has a huge tie to CoreWeave, and some people view that as a potential kind of backstop for CoreWeave in a way. So some people view that as a safeguard, and other people view that there could be a lot of risk in how closely the two companies.
Speaker 1: are intertwined. When CoreWeave and NVIDIA announced that investment, I was interviewing CoreWeave CEO Mike Entrader and NVIDIA CEO Jensen Huang, and I got to ask Jensen how he feels about the circular finance as it relates to this exact deal. Suffice it to say, Jensen Huang does not agree with the circular finance claim. His point was, well, you know, we're giving CoreWeave $ 2 billion. They're spending far, far more than that on NVIDIA chips. So it's, you know, but to to that point still, you end up with a situation where maybe it's not covering the whole chip bill, but NVIDIA is handing CoreWeave some amount of investment and CoreWeave does turn around and spend that on NVIDIA chips. And I think the circularity concern here comes from how intertwined everyone is. There's some element of backstopping in the deal where if CoreWeave can't find customers for certain GPUs, NVIDIA maybe helps either find customers or uses them for some of NVIDIA's own partners. It is very interrelated. So that becomes a concern here. You asked also about other significant CoreWeave shareholders. Jean Street is a significant CoreWeave shareholder and customer. Magnetar is a significant CoreWeave shareholder. And so those are companies that also have significant exposure here if this jack goes down.
Speaker 2: Another headwind I've started hearing a little bit more about is this idea that companies are limiting token use for employees. They're engaging in more token optimization. Is that something that a neocloud like CoreWeave might find challenging?
Speaker 1: I don't know that that's a huge deal for CoreWeave based on who their customer base is. I think they're still using quite a lot. I think what you're seeing at companies is, one, there's some amount of people just running up their token use to show their managers that they are using AI and therefore they're a good AI employee. I think it's part of the ongoing question of AI usage and demand has continue to grow at a pretty exponential level. At what point does it slow down and then what happens? Is there any overbuilding? Is there any overbuying of certain chips? Do we end up in a situation where certain chips are less favored than others when you have the ability to choose between them? Right now, there's not enough of anything. People are buying what they can get. If that changes, there's a lot of economic decisions that are different.
Speaker 2: Yeah. The and then what happens question, I feel like, is one that comes up every week and we still don't have an answer.
Speaker 1: It's like that in every massive shift in infrastructure build out, right? I mean, it's the and then what happens followed by and who's left holding the bag.
Speaker 2: This is The Big Take from Bloomberg News. I'm Sarah Holder. To get more from The Big Take and unlimited access to all of Bloomberg.com, subscribe today at Bloomberg.com slash podcast offer. Thanks for listening. We'll be back tomorrow.