Where you see this tension between execs, it's usually a reflection of how pressured things are behind the scenes.

This is not what big companies that are about to list for one of the biggest IPOs ever are supposed to do.

I think Oracle is a bit of an outlier because they are investing in AI infrastructure.

Actual bankruptcy isn't a huge risk, but what is a risk is they say, "Yeah, we signed this long-term contract, but we don't have the cash to pay it right now." We're going to talk about project finance, how this machine is funding hyperscaler investments.

This is in advance of the IPO even, right? Because they haven't even actually filed for their IPO yet.

You've got Sam Altman and their CFO having this public debate about an argument about missing their revenue number.

Welcome to this week's episode of "What's the Big Deal?" Where we do talk about major deals in the public and private markets and explore finance industry developments. My name is Deborah Taylor, and I'm going to use my background from my career in investment banking to talk about things from a public markets perspective.

And I'm Graham Smith. I'm going to use my background in private credit to talk about things from a private market perspective.

Fantastic. Right. Graham, let's dive in.

What is the big deal this week that we're going to talk about? So, no surprise, we're back to-- obviously, we got a few big recurring themes this year with AI. We've talked about this in the context of a few different things, and actually, it's kind of really, it's obviously topical because there's a lot going on.

We wanted to have kind of an educational episode today on hyperscalers and kind of perfect timing in the sense that OpenAI just missed a revenue forecast. They're obviously one of the big hyperscalers, really taking a lot of data center capacity.

So, both want to have a chat about what's going on with OpenAI, talk about the overall landscape, and if things are wobbling a little bit, then talk about just some early views on who's at risk and what might happen over the next year or so.

Absolutely. So, we've got you starting off on the numbers this week. You'll talk about why a single missed revenue number is actually affecting the public markets.

I know we've had a big sell-off on some stocks over just the last couple of days.

We're going to talk about project finance, how this machine is funding hyperscaler investments, and who's funding this and how. And then finally, we'll have a little discussion about why data center investments are looking maybe a bit more risky today than they were even a few weeks ago, and whose risk is that? That's the really big question, isn't it? Graham, the news on OpenAI's revenues, why does this matter? Well, this actually follows on from some of the discussion we were having on the OpenAI Anthropic IPOs that are coming down the track this year. And if anyone has listened to that episode, one of the things we talked about was just some of the hyper-aggressive revenue recognition policies that look, we don't know for sure that they're using, but you can kind of venture an estimated or an educated guess here and really say that, one, you kind of know a lot of these high-growth tech companies in the run-up to IPO are going to recognize revenue as aggressively as they possibly can.

Look, I don't know what your view is on OpenAI, Sam Altman, just the whole cast of characters involved in this overall landscape, but just knowing what you know about them as individuals, do you think they are being conservative with their revenue recognition policies or perhaps a little bit aggressive or slightly too aggressive? So, I think when we see some news that the revenue-- This is in advance of the IPO even, right? Because they haven't even actually filed for their IPO yet.

They've missed a revenue target, and they've missed it based on these super aggressive revenue recognition policies that I think we all assume they're using. And you also look at the other stuff that's happening with OpenAI in particular, and you've got Sam Altman and their CFO having this public debate about an argument about missing their revenue number. And I think that's just made a lot of people say, "What's going on? This is just crazy. This isn't what companies are supposed to do.

This is not what big companies that are about to list for one of the biggest IPOs ever are supposed to do." And I think you put that in the context of you've got all this committed CapEx spend across all these hyperscalers, and people are saying, "All right, if we're already missing revenue targets, what is going to come of all this data center capacity that's already been committed and is both built already, coming online, all the committed CapEx that's been made in this space?" I think everyone's just taking a step back and saying, "Wait a minute, what's going on?" Yeah, because it does sound incredibly chaotic from their side.

And definitely, we've seen examples of this before where you see this tension between sort of execs, it's usually a reflection of how pressured things are behind the scenes. So, that in itself for me is a very, very poor signal.

But you mention around the commitments, because we're talking here about revenue misses, but they are committed to huge amounts of spending off the back of quite aggressive growth targets, aren't they? Whereas, we're normally thinking about- Yeah ... OpEx against current revenue numbers.

But because the growth of revenue was expected to be so aggressive, they've signed up to big future spend that they can't easily walk away from, can they? No, and I think the other thing is you have these big, when we say hyperscalers, we're really talking the huge tech providers, like the AWS's and Microsoft's and Meta's and OpenAI. And I think everyone, we'll talk a little bit about project finance and everyone who is actually building and financing these data centers. When you have a contract signed by one of these big hyperscalers, you kind of think of that as pretty bulletproof, and you secure equity financing, you secure debt financing against these long-term contracts. And then when you have one of these really big guys say, "Oh, I don't know if I've made my revenue target," I think that makes everyone freak out a little bit, because to your point, they've committed just a ton of CapEx spend off expectations of hypergrowth. And if that growth doesn't come, then the question is what happens and who's exposed? Hmm. Yes, it is a big question. What happens and who's exposed? We'll definitely get onto that.

Yeah.

So should we talk a little bit about the financing of the CapEx spend and sort of how that's being financed? Because these are big numbers that we're hearing.

Yeah. And so, the numbers, so I'll kind of run through just the high level numbers, and then you've got more project finance background, so I'm actually really keen to get your take on just how all this stuff comes together.

Hmm.

All right, so I'm just looking for, I think this is just for this year.

We're looking at a total committed CapEx spend of 700, 725 billion across some of the main players.

So Amazon, Microsoft, Google, Meta.

Amazon, 200 billion, Microsoft, 190 billion, Google, 190 billion, Meta, 145 billion. I think literally as we've been looking this up, we've seen some press releases, I think it's Meta in particular, even upping their CapEx target. So everyone is just going all in on data center build, compute capacity, and it's really driven by these big hyperscalers. The other one that's kind of interesting, and I think you know a bit more about this than I do, is just how Oracle fits into all this, because it's actually kind of interesting.

Oracle has been around forever. I feel like until a few years ago, people probably forgot about them a little bit.

Hmm.

I mean, it started as a database business and it's ERP software, which, I don't know, is not particularly exciting.

And then they really jumped in on the AI bandwagon, and to date, have been real winners. But it also sounds like they're potentially on that list of parties who's pretty exposed here.

Hmm.

So that's literally just for 2026, the better part of a trillion dollars of committed CapEx spend on data center compute capacity. And admittedly, when we say spend, that's not just on the physical buildings. That's on the servers, the GPUs.

Obviously, the GPU is a massively expensive component of this CapEx.

Absolutely.

And I don't have the details of how all this breaks down necessarily, but the numbers here are big.

Yeah, because I think everybody wants to be on that bandwagon.

There have been big numbers thrown around.

Everyone's talking about AI investment, but without specifying what they mean by that investment. Is it data centers? Is it chips? And also, it's not always clear how all the pieces of the jigsaw fit together.

You mentioned Oracle, OpenAI.

There are other people who are involved in this AI investment, but sort of how does it all kind of connect together? So I think we should talk about that a little bit, because that kind of, I think, gets to the nub of why this OpenAI news is so important.

Because potentially, it kind of is the first link in a whole chain of things that could potentially unravel if worse comes to worst.

But let's talk about the financing for a lot of this spend.

So in terms of the financing for all of this huge CapEx spend, it is coming from the debt markets, and roughly half of that is coming from the public markets, so that's like bonds, and half is coming from the private markets. Now, I think on the public market side, what's happening with bonds is really well talked about.

A lot of us have heard about all the bonds that have been issued by the hyperscalers, Meta, Alphabet with their 100-year bond.

There's been a huge amount, $1.2 trillion of bonds have been issued over the last, up I think up to now, in terms of funding AI spend. And that's spend that's coming from the hyperscalers. It's on their balance sheets.

Now, a lot of this is being deployed in purchasing chips.

Those are the GPUs, which you explained so beautifully in a recent episode.

I think Oracle is a bit of an outlier because they are investing in AI infrastructure. They have a much weaker credit rating than the other hyperscalers, so they can't tap into the bond markets quite to the same extent as the others. So they've actually arranged some structured financing. I think it was led by Pimco recently. They basically arranged some financing led by them. So we have got some spend which is actually on the corporate balance sheets, but I think what's maybe less well understood is the element that's being provided by the private markets through project finance. And project finance isn't a new thing.

That is a decades-old form of financing infrastructure projects.

Yeah.

And for example, the Channel Tunnel, power stations, all of those are funded through project finance, but now it's being repurposed for data center financing.

And the way that that works is, I think, quite interesting in kind of peeling back the layers of how different funders are involved. So how does project finance work? So first of all, a special company is set up, what we refer to as a special purpose company or an SPV, to construct and operate the data centerAnd you might have heard of SPVs after Enron. They got a bit of a bad rap, but they're widely used for arranging these kind of structured finance arrangements.

Now what that SPV needs is really predictable cash flows. It needs to be really low risk in terms of what the cash flows are going to be for the construction, and then the cash flows that will come in once that data center is operational.

Okay? So what the SPV does is just signs loads of contracts to take the risk outside the vehicle.

So that would be like construction contracts, so a fixed fee contract for constructing the data center, and then revenue contracts for when it becomes operational. Now, those could be leases or what we refer to as take-or-pay contracts, where the customer, if you like, for the data center agrees to take and has to pay for that, the compute power from the data center, even if they don't use it. They're locked into that contract, and that is actually really important- Okay ... in terms of when we're thinking about the OpenAI effects. Okay? Now, once you've done that, once those contracts have been signed, you've now got really predictable cash flows, and that means you can now use loads of debt financing to fund the construction side of things. And when I talk about lots of leverage, we're talking more leverage even than you were referencing in last week's episodes around leverage buyouts.

We're talking about 90% debt financing, 10% equity financing.

Right. So this is extremely leveraged. But on the basis that the cash flows are going to be very, very predictable. Okay? So basically what you do, just for my own benefit and for everyone else- Mm-hmm ... you set up this SPV, you go out and you arrange everything before you start breaking ground on the actual data center.

So arrange the construction, the contracts, right? And then once everything is kind of signed and delivered, then you go out to the financing market and say, "Okay, I've got this just insanely robust set of lease payments that's signed up for..." What, I don't want to-- A typical lease term is on a data center like this. I think 15 years or something, yeah.

Fifteen years. So and I assume that extends beyond whatever kind of credit facility is extended or offered to these SPVs? Yes, I'd assume so, yeah. But the idea is that the revenues, if you like, for the SPV, they are going to basically be used to pay down that debt in a similar cash flow- Right ... waterfall to those that we see with leverage buyouts.

So as soon as the revenues start coming in, you start paying the debt that was raised to fund the construction. So that's the way that things are set up, basically.

Is there any element of-- Because when we talk about 10% equity, it reads a bit CLO in structure. Are these the kind of things that have different tiers of debt, in essence? Or is it just one big facility and whoever participates takes the same return? Yeah. I mean, I don't know the layers of financing in as much detail as I do for- Yeah, okay ... leverage buyouts. But yeah, there are different tranches within there.

I don't know what a typical- Yeah, okay ... structure would look like.

Right.

But yeah, it is very leveraged and ultimately, it's done on the basis that you have very predictable cash flows. What makes it really interesting is that from a tech company's perspective, they're signing these contracts with the SPVs, and it doesn't actually go on their balance sheet.

It's effectively off balance sheet until the data center becomes operational. So it's attractive from their perspective.

It's attractive for those investing in it because they're basically lending to or providing equity finance to a vehicle with very predictable, stable cash flows. Or that's what you anticipate. Now, a key- Yeah ... part of the SPV is that it is a bankruptcy remote vehicle. That means that those that invest in that vehicle have no recourse from the hyperscalers. So if things go wrong, the lenders can't tap into the hyperscalers unless there's some kind of additional guarantee that's being provided. Okay? And this becomes important if things do start to go wrong.

However, we also know that the hyperscalers, if they have lease commitments or even take-or-pay contracts, they can't walk away from those either. So we really then need to explore how this chain unravels if the party at the end of the chain, that's OpenAI effectively, if they can't afford their commitments, how does that then affect Oracle, who basically rely on them for their revenues, for the compute? How does that then affect the SPV that's basically got the contract, the take-or-pay contract or the lease? And then how does that affect those providing the finance to the SPV? We've now got a weak link in the chain. How do things unravel? Ooh, there's a lot of stuff in this chain to get your head around. So let's just play out a quick scenario and say, okay, it's probably unlikely that any of these big hyperscalers is going bankrupt.

I guess you never know with a company like- Mm ... OpenAI, right? Never say never.

But let's assume for a second that actual bankruptcy isn't a huge risk. But what is a risk is they say, "Yeah, we signed this long-term contract, but we just don't have the cash to pay it right now. We missed our revenue targets.

We're already burning literally all the cash in the world, so sorry." What happens in that scenario? A key issue here is that two-thirds of OpenAI's compute commitments are to Oracle.

So Oracle has a big exposure to OpenAI, and that, I think, really means that we should focus in on Oracle's kind of position or role within this chain. Yeah? Right.

So let's say OpenAIThey disappoint on revenues. Again, they're now in the position where they're going to miss payments to Oracle, let's say.

I guess the first question, Graham, you probably have more experience with this than I do, but can things be renegotiated? Could they even raise additional equity? Those are the first steps, aren't they, when you're thinking about restructuring.

100%. I think that the answer is things can always get renegotiated, right? When things start getting into trouble, as a lender, you're generally not just hitting the button on your full-on enforcement on taking the keys to the business goodbye kind of process.

Yeah.

There's a discussion to be had around some kind of compromise.

So you're right, I think that's- Yeah ... probably what happens.

Yeah.

But it does feel like Oracle perhaps right now is the most exposed in this chain. Is that fair to say? Yeah. I think it's a big if, but if OpenAI genuinely can't pay, can't renegotiate, Oracle does become the next focus because they are locked into those SPV contracts.

They're either lease contracts or take or pay contracts.

They can't walk away from them. And as I mentioned earlier, they are the only one of the five big hyperscalers that don't have a high or a really good investment grade rating.

They're just above junk status. And I think- Right ... their credit rating would be the next domino to fall.

And unfortunately, that would mean them being downgraded to junk status, which isn't just a change of label, it can be a covenant breach, for example. So for them- Yeah ... it puts them under a lot of pressure within the links of the chain.

I think in reality, if then they can't pay, then the SPV is left, of course, exposed, and those that have invested in the SPV are also exposed. But in theory, you would then expect that they could find a new tenant, I think they refer to them as anchor tenants, the hyperscaler that's kind of going to be using the data center. So it could be repurposed, presumably, to a new anchor tenant.

And presumably, so when we talk about, and we don't have the split, but when we talk about this $600, $700 billion of committed CapEx spend, that is both the data center build and all the stuff that goes inside.

Presumably, the SPV is not really on the hook for the stuff necessarily if the cash doesn't come in and the data center build doesn't finish, or are we in this world where you finish the data center build, you get the client in, they stop paying, then I guess there's a bunch of other financing for the stuff as well.

Is that all the same SPV? Is that different financing parties? We've been hearing about ABLs on GPUs.

Mm.

I feel like the number of counterparties here is almost too big to keep track of.

Yeah. So I think, definitely there's kind of a parceling up of the different assets within the data center. As you say, there's the GPUs.

It has its own separate financing.

But the actual physical building, all the equipment within there, I think is all covered by these contracts, which are so important to how the SPV functions.

So once that's been signed, you're locked into that construction, and therefore you're then locked into operating that data center.

And I think that's the big issue for the SPVs is can they find a new tenant? There is a small chance that you end up with what we refer to as stranded asset risk, where you have an asset which actually doesn't have really much economic value, maybe because technology has moved on in the few years whilst the asset was being constructed.

And it's fine if you've got a tenant who's locked into the contract, but if you're then trying to find a new tenant, but that asset's now aged in terms of technology, could it become a stranded asset? And I think that is a scenario where the investors in the SPV would be exposed. So there's, I think, a lot of ifs.

Yeah.

But maybe- Yeah ... fewer ifs around Oracle's exposure.

Interesting.

Are we in a world in which the Paramount-Skydance-Warner Brothers deal is now off because all of a sudden Larry's like, "Sorry, David, I don't have enough cash anymore." Oh my goodness.

"I got to make good on all my data center commitments." The chain keeps unraveling.

Wow. Where does it stop? But look- Seriously, I kind of take your point on the stranded asset thing, though, because I guess if you believe that we're in a world where OpenAI, as an example, can't make its data center payment, then who else gobbles that up? If the hyperscalers aren't able to satisfy the demands, then who is? Yeah. And I think so far a lot of the risk has been about whether there was a risk of oversupply in the market.

There's been so much investment in the infrastructure.

And I think it was just talked about in very hypothetical terms, and now we're actually really starting to see that this is a genuine concern, and having to understand the different pieces- Didn't take long for that, did it? Well, I think we're two years in, but yeah, I think we now have hard numbers, or at least, well, it sounds like they're quite soft numbers, to be honest, for OpenAI, but we have some data- Yeah. Seriously ... to work with, because a lot of this has been on the private side, with no real numbers being released until recently.

But I think it is interesting because we've talked about this as being OpenAI's issue, but it's not really just an OpenAI story. It's about how the financing of AI infrastructure has kind of grown and assembled itself over the last couple of years, and it's now effectively being stress tested.

Now, can it cope with the risk that you have maybe one company at the end of the chain not performing as expected? And you know-Ideally, if the people structuring it have done their jobs well, and the risks have been well managed, it shouldn't unwind in too chaotic a fashion. But who knows? Yeah.

And it's definitely a story where we had so much upside. It's all been positive, hasn't it? All the messaging.

Invest in AI. Anyone who was mentioning AI investment in their press release, it was seen as a big positive.

Is the tide turning, I guess? Maybe, and we'll see what happens in coming months and coming years. Because the one thing that is true in this space is things change so, so quickly. And if there's, let's say we get to a point six, 12 months down the line where, because the new compute capacity comes online, we see AI models that are just so much better even than the ones we know now, then are we in a position where all of a sudden revenue's growing again, and then we get into bigger discussions about everyone's losing their job because AI's taking their job because it's so good and all the things.

So there's just a lot to unpick here.

Yeah.

And no doubt we're going to keep talking about this just because it's interesting, it's topical, it's moving quickly, and in some ways, we've been talking about the financing today, but also it impacts everyone to some extent.

Mm-hmm.

So it's a highly interesting topic.

It is interesting. I don't know about you, but I feel like every day I'm learning more about how all of the different pieces work. I think a lot of us have used AI without really questioning how the companies operate, how the financing works, and so it is a good opportunity to learn a little bit more and open your eyes to what the future is potentially going to hold.

100%. Because it's both bright and also quite dark at the same time, depending on what you take and where you think it might go. So lots of stuff to think about.

So I think we've covered quite a lot of ground there in this week's episode.

We'll take a pause now for this week.

But thanks ever so much for those of you that have been listening.

And if you're watching on YouTube, please do like and subscribe and leave us a comment.

And if you're listening on Spotify or Apple, please do like and leave us a rating.

So thanks very much from me, and over to Graham.

And thanks from me as well. No doubt, lots more to come on this topic.

It's going to be a busy year. We still have the big IPOs to come. I think the interesting thing for me is that we're going to get a lot more actual data and have some numbers to really talk about once we see some investor reporting come out.

So watch this space, because there's going to be a lot more to talk about.

So it's been fun, as always. Thanks, Debs, and we'll see everyone same time next week.