Bản chép lời đầy đủ của Syndication Nation — From mega-cap to megawatts: How AI data center bonds are rewriting the credit playbook

William Hoffman They're also bringing together teams from, you know, structured analysts and commercial real estate experts and legal teams. Like, all these investment firms have those teams and they're leaning on those different teams. But it's also how can they create a sort of one-stop checklist for how these things get measured? Sunny Oh And so, I think, you can check our article, but I think that our conversations on the street, we compile, this crazy long checklist that analysts are using to grade these bonds. And I think it spans five categories? Like, obviously, we can't go through every little question, but could you just give me a general rundown? William Hoffman Yeah, not to go through everything, but this was a common theme that kept coming up in all the conversations that we had for this article, which is quite a few, actually. Which is, what are the questions that people are asking through? So the big categories are construction, power, leases, structure, and fundamentals. Just kind of briefly go over each of those. Like, construction, who is the construction company that's working on this data center? Are they reliable? Do they tend to have overruns? Are they IG rated? Are they high yield rated? So, those are some of the things that we're looking at there. Power: This is a big one. These are essentially boxes, big boxes that are hooked up to utilities, right? There's not much going on inside of them besides that until you fill them with the tech, right? So, where's the power coming from? Is it on-grid? Is it off-grid? Can it be displaced? These are some of the things that people are asking there. And it's a huge part of that process. Leases: We’ll get into kind of some of the details here. But the lease agreements are very different across a lot of these deals and makes a big difference about whether it's IG rated or high yield rated. Structure: You're looking at some of them are amortizing, some of them are not. You know, are there exits for the guarantors? Where are the cash flows coming from? And then, finally fundamentals. These are the questions that I think a lot of people are having at dinner tables, right? Is this an AI bubble? Are tech companies overspending on data centers? All those big questions kind of go into that bucket. And it's kind of why these decisions are centered around the tech analyst. Because once you make your way through the checklist and down in this fundamentals part, that's where the specialty lies. Sunny Oh Okay, interesting. So, I think we'll maybe start off with how this checklist might apply in, say, high yield. Just because that's where the market kind of took root. So, I think we tracked some of these deals, I think, from my understanding, there's 17 deals where the collateral is tied to data centers. The cash flows come from the lease payments. And I think 13 of those deals are rated. So, in high yield, that entire credit risk is driven by who the tenant or the lease guarantor is. William Hoffman Yeah, and the yields can tell, like, a very clear story here. The ratings can be quite different. But when you look at the yields, you can certainly tell which ones are lower, which ones are higher, where the risk is being placed. Sunny Oh Yeah, no, definitely seems the case. Like, I think we look at Applied Digital. I mean, I think that is the perfect example. They could price two separate bonds, one backed by CoreWeave, another backed by Oracle. Very different names. But at the end of the day, both bonds had a very similar double B rating. However, I think the bond backed by CoreWeave, when it priced, came with a nine handle. While the one backed by Oracle priced at high sixes. So that's a massive yield difference, right? So, it shows exactly how much more risk investors are assigning to one of these newer cloud providers that came from the Bitcoin mining space like CoreWeave, compared to one of these kind of blue chip hyperscalers like Oracle, you know? I think one source told us deals with these hyperscalers they price the low 6% area. While non-hyperscaler deals could trade maybe, like, 200 basis points wide. William Hoffman Yeah, but the CoreWeave backed deals, right, those have kind of been tightening lately, haven't they? Sunny Oh Yeah, and that's fair. And I think that also speaks to this enormous demand for these kind of offerings. People cannot have enough of them. So, and I think that's also because we saw a recent headline where Jane Street, a quant, signed a massive $6 billion AI cloud agreement with CoreWeave. And I think that's, seeing these headlines, these massive contracts, they push these CoreWeave data center deals right into almost the 6% trading area now in the secondary. But the structural features do create divergences in how these deals trade. So, it's not like they're all trading at the same level. I think it's we might look at something like amortization schedules. So, if you look at Fleet Data Center, I think they had a $4 billion bond due in 2031. Only around 7.5% of that principal amortized by maturity. So, I think at the end of the day, that leaves a $4.2 billion payment that needs to be refinanced at the end of the bond's life. So, you know, it's a huge amount. William Hoffman Yeah, it's a large amount, especially for, you know, a smaller provider that's not, like, one of these, like, hyperscalers that has the balance sheet. Sunny Oh For sure, for sure. And I think compare that to, say, like, Cipher Mining. I think there’s a $1.7 billion bond, 2030s. I think nearly half of that bond fully amortized by the time it matures. So, because of these slower amortization schedules on certain deals, I think a lot of the high-yield PMs might look at these structures as shorter-term two-year paper that will need to be refinanced, ultimately. William Hoffman Yeah, absolutely. And, like, that's the high-yield sector, right, where a lot of these deals have started. I think those are getting a little bit more

standardized, right? I think when you switch over to investment grade we've kind of seen that those deals, frankly, just haven't had the standardization across the structures that the high-yield spaces had. The biggest takeaway for corporate credit investors is that a high-quality tenant alone isn't enough to drag a project into investment grade. These structures typically need to have much tighter lease terms or an ironclad construction cost guarantee to get them over that IG hump. Sunny Oh Interesting. Okay. And I think you did a great case study in our reporting, right, comparing, Meridian Compute and Hut 8. I think those are the two newer issuers in the IG market? William Hoffman Right, yeah. Both deals featured Fluidstack as the tenant there, and both are backed by guarantees from Google. So, two data centers that have the backing of Google, essentially. Sunny Oh So, it should be like for like, almost. William Hoffman It should be like for like. And yet, Meridian is rated high-yield with mid-BB ratings, with Hut 8 getting investment grade at BBB minus lowest tranche, right? The difference was structural in de-risking. Hut 8 brought in a higher-quality construction partner that explicitly guaranteed a maximum cost to build. Meaning that, if they went over, the construction company was going to pick up anything and that was not going to fall on Hut 8 or Google, right? Sunny Oh Mm-hmm. William Hoffman So that boosted it up. And furthermore, Hut 8 guarantees consistent lease payments if Google were to leave. Or I guess Google is guaranteeing those lease payments. Whereas in that Meridian deal, the lease payments would step down. You're not guaranteeing those lease payments. So it's kind of a a good example that you can have similar deals that a few structural tweaks can really make a difference. Sunny Oh Interesting. And I think we have it recently that QTS deal backed by Microsoft. That was a big hot topic in the market. It seemed like it gave investors a big headache, and I think we saw that in the trading, too. William Hoffman Yeah we really did. That was a different deal because, mainly because it's not amortizing. They're not amortizing at all. You're basically relying on the fact that Microsoft, at the end of this lease term, is going to pick up the full cost of the bond. Sunny Oh Mm-hmm. William Hoffman Which is very atypical especially for a special financing construction bond. But because it's Microsoft, there's a lot going for it. So some of the counterweights. This would typically be a one-way ticket to high-yield if you were doing a non-amortizing bond. But a few things working for its favor. One, the data center is already built. No construction risk. I don't know if it's running, but the construction is done. Second, it's backed by Microsoft, like we said. So, AAA-rated credit. I think the last AAA-rated credit. They haven't issued since 2017. There's a lot of demand pent up for Microsoft exposure. Stronger than treasuries, some would say. And then finally, Moody's highlighted this in their report: tenant stickiness if this facility. It's a data center that's synchronized across Microsoft. It' two facilities, really. And it's part of their large language model business. It seems like it's sort of a flagship for them in the data center space. And you'd need something to really wrong to happen in the AI space for them to abandon this project. Sunny Oh Mm-hmm. William Hoffman So all of those things worked in its favor to give it IG ratings. Even so, without that amortization, investors are taking a long-term five to fifteen year bet that Microsoft will stay dedicated to AI data centers. And that's based on the premise that space that this will remain a cash cow for Microsoft. Can I get really technical for a second with you? Sunny Oh Yeah. Yeah. William Hoffman Cool. If you want to really dig into this, right? The rent, I believe there was a mechanism that Microsoft has to pay off all the rent payments to the end of its 20-year lease agreement, even if it decides to leave, right? Sunny Oh Mm-hmm. William Hoffman So depending on when Microsoft where to theoretically leave, that would allow those future rent payments, lease payments, to cover the principal or not. So if you're five years into the lease, and they decide to leave, then there's probably no issue because there's probably 15 years of rent payments that they would be guaranteed to pay out and that would cover the principal. But if you get to the 10-year mark, this is only a 10-year bond, you have to refinance at the 10-year mark. And then, you're five years into the next refinancing and it's not amortized, then that next five year period wouldn't be enough to cover the principal. So this is kind of a long-term risk play. And who knows what's going to happen in 2041. That's 15 years out. That's the make or break point, and no one really knows what's going to happen at that stage. Sunny Oh Yeah. I'm sure. I'm sure a lot of the guys who are investing in these things won't be there to see what happens with these deals. William Hoffman Right, exactly. Sunny Oh Exactly. But so do you think it was worth it for these issuers to add these extra guarantees and restrictions on the debt to get these IG ratings? William Hoffman Yeah. I mean, I guess it depends on, you know, the perspective of the issue they were able to finance it at cheaper rates than some of the other deals in the high yield space or private credit space. The IG deals mostly priced in like the low 6% area or even like high 5% in the case of QTS. By comparison, most of the high yield deals were had coupons in the high 6% area all the way up to 9%. Although some of them have traded down lower than that since they've priced. But that jump from the upper rungs of high yield to the lower rungs of IG is only saving some issuers 50 basis points. If you're Google, like you're not really sweating 50 basis points of interest expense, especially when it technically sits

off balance sheet through these data center developers. It seems like most would prefer some flexibility in the event that AI is being like overhyped, and all this is just a bubble. Sunny Oh Interesting. OK. So what's wild to me is that some issuers are actively choosing to stay in the high yield market and pay up for the private market execution, even if they could aim for IG. I mean, I think you sort of explained it pretty well there. But I mean, we saw Meta and Oracle finance sales in the private markets and then syndicate a portion out of that into 144A debt. You know, it's kind of like this these interesting like hybrid public private offerings and they paid higher coupons of what they could have like, in theory, achieved in the public IG markets. What's going on there? William Hoffman Yeah. I mean, it's all about the premium for flexibility, right? So, you know, a good example is this Meta deal. It really kicked off these sorts of data center financings last year. It was a $27 billion private transaction, issued by Blue Owl because they were providing some equity in the data center as well, but then backed by an $18 billion anchor order from Pimco. These structures were so new to the market last fall that they just opted to pay for the certainty of execution with a few select players that really knew the structure and they could really talk to and engage with. That deal was for a data center based in Louisiana. So they named the bond deal after the local beignet pastries. Still really confuses me every time I look it up because they use the French spelling instead of like the Americanized spelling. And it throws me off every time. So now Meta is actually reporting reportedly looking to fund its $13 billion Texas data center and a bond deal named after another local pastry, the sopapilla. Sunny Oh Well, what is that? William Hoffman It looks tasty. I've never had one, actually. Maybe should go to Texas and try it out. But, you know, I think that next deal will be like a really telling test of the market if they go back to private credit. Or is this a time to look to kind of the broadly syndicated market and see if they can get lower coupons now that the market is more aware of this kind of structure? Sunny Oh OK, yeah. And I think, based on our conversations, though, it seems like asset managers are getting frustrated because they want more standardization, but just give the nature of what they're financing. It might be a tough ask. And maybe you don't want it to because there's a bit of a structure in premium, too. If it was very standardized, if it was all very plain vanilla, you wouldn't really get that yield. So it seems like every project has a unique leasing and power constraints and companies don't want to give up their like optionality just in case just to standardize something like they wouldn't really make sense for them. Right. William Hoffman Yeah. The investors want it and it would make going through these deals a lot easier. But if they eke out some extra flexibility or customize it to the specific site, they're going to do that. And that’s typical of what this is, which is project financing. That's why people are specialized in this sort of trading, because it's a very specialized sort of thing. And they're trying to bring it to a more standardized sort of way. And people are, feathers are a little bit ruffled from that, I suppose. In the meantime, I think the pipeline is going to keep coming. It's absolutely roaring. In the high yield space we hear there's like another $6 billion of data center bonds coming down the pipeline. That's probably coming to the high yield bond market could come construction loan market. But either way, like a lot of these deals still in the pipeline. The Meta deal we just talked about is another multi-billion dollar deal. And issuers love the speed of the bond market over project financing or construction loan space. Deals can get executed in as little as six weeks, which is like twice as fast as a project financing deal. There's also, we only looked at deals here that are backed by data centers, but there's also some other like weird maneuvers you could do. Like CoreWeave did a high yield deal that is tied to the GPUs that go into the data centers, right? Which is not, it's spread out across different, you localities. And that's a different sort of function than the data center and the utility hookup to the box. But it also like a different mentality because how do GPUs perform over time? Do they degrade? Whereas like the utility hookup to a box is pretty standard for the most part. So, yeah. Anyways, there's a lot of, there's going to be a lot of different variations of this kind of style. But this structure is like going to be very popular when you need to raise $3 to $5 trillion through 2030. Sunny Oh Yeah. No, no. And anecdotally, that CoreWeave deal, it did amazing in the term loan market just because there wasn't anything like it. And I think it just shows that there is just this avid demand for investors who just want to take advantage of this AI boom. You know, maybe some of their part of the portfolio doesn't benefit from this. And it seems like primarily when these things were first marketed, that was the appeal. You're going to have that, who knows what's going to happen in the world and in AI, but in case this becomes a big thing, you are going to benefit from it. It seems like that's the pitch, right? William Hoffman Yeah. And Google and Alphabet and all these hyperscalers, they've been sitting on mountains of cash for like years and investors have been, you know, begging them to issue more debt. They want exposure. Sunny Oh Yeah. William Hoffman And now they're getting it, you know? Sunny Oh And so you can't complain. William Hoffman You can't complain. And these data center deals are the highest yielding of them. They're frankly the lowest rated, highest yielding. And it's a great way to get high yield exposure to what are like ironclad, A-rated tech names. Yeah. So still early days of the market, at least from what people are telling us. And we haven't achieved like a standard checklist of like what people should be asking. But, you know, I guess until then, investors will have to keep sifting through this complexity if they want to get a piece of this fast growing market. So that's all the time we have for today, but thanks for tuning in and thanks, Will. Hey, thanks for having me on, Sunny. Sunny Oh Yeah. So still early days of the market, at least from what people are telling us. And we haven't achieved a standard checklist of what people should be asking. But, you know, I guess until then, investors will have to keep sifting through this complexity if they want to get a piece of this fast growing market. So that's all the time we have for today, but thanks for tuning in and thanks, Will. William Hoffman Hey, thanks for having me on, Sunny.