00:00Joining me now from the Carlyle Group's annual Global Investor Conference is Carlyle Head of
00:04Global Research and Investment Strategy, Jason Thomas. Jason, thanks so much for sitting down
00:09and for having us here at your Investor Conference. Great to be with you. So we are just over 24
00:13hours
00:13away from a Fed decision. And what a backdrop we're seeing with Brent crude over $100 again,
00:1910-year yields surpassing 5%, the highest level in nearly two decades. How are you thinking about
00:24that backdrop as it applies to what we're going to get tomorrow? I think the Fed's under enormous
00:29pressure to deliver a 25-basis-point hike, I think that's not just the front end of the curve.
00:35I think it's also from the electorate. If you think about this year, the primary campaigns,
00:41almost all waged on the issue of affordability. Many victorious candidates ascribing the dramatic
00:48cumulative increase in prices to market failure, recommending massive state intervention to
00:55remedy that. And I think that that has changed the lens through which we think about inflation.
01:01You know, very often it's this technocratic issue, thinking about this category, that category. Well,
01:06if you take this out, inflation's only 2.2%. But there's a broader political economy here.
01:12People have been hurt by this cumulative increase in prices. The living standards have declined. And I
01:19think the Fed has to get serious about delivering on the price stability mandate.
01:24That's really interesting because often this decision has been framed of it's happening right before the
01:29midterms. It's something both the president and Kevin Hassett, his economic advisor, have talked about.
01:33This isn't a Fed. That should be hiking rates before an election. It kind of sounds like what you're saying
01:38is the inverse, that actually it's politically advantageous to be hiking rates in order to help cap
01:44inflation. Is that right? I think that there's this broader political discontent that very much
01:51can be tied to the fact that prices, the price level in the United States has increased 27%
01:57cumulatively since the first quarter of 2021. So this is not great timing before an election. If it
02:04could be avoided, that would be great. I think in most cases, the Fed would look through a supply shock
02:10delivering the inflation as we see today. But these are not normal times because of the cumulative price
02:16increases we've already observed. What does that say about whether this hike, which markets are
02:21pricing at more than 90% odds, is a one-off or the start of a rate hiking cycle? It'd
02:27be unusual for it
02:28to be a one and done. Of course, it'll be data dependent. But I think most market participants should
02:34expect that the Fed will perhaps hike again before the year is out. There's two more meetings. I think
02:41an additional hike at one of them is probably more likely than not. Well, in the meantime,
02:45as I mentioned, you've seen yields climb above 5% for the 10-year for the first time since 2023.
02:51At what point does that get to a level where it causes more damage, be it economic,
02:56be it in risk markets? Because for now, it has been an equity market that's held up remarkably well
03:00in the face of rising yields. You know, I don't have any specific number in mind. But I do think
03:07when we look historically, there are parodies between, of course, real interest rates and
03:12earnings yields. And as those real interest rates get higher, it starts to become advantageous to take
03:20a bit of risk off of your portfolio. And so when that's actually triggered, who knows? I mean,
03:27I think as long as there is this embedded optimism about AI and about future earnings growth prospects,
03:34as long as earnings continue to be as strong as they've been, there's still going to be a lot of
03:39risk appetite. Can we talk about AI? Because you've just released a really, really fascinating paper.
03:44And part of the core of the thesis is you talk about what's priced into this market now. That is
03:48an engineer's view of AI. That in order to advance, you need more compute. So the CapEx makes sense.
03:54Is that the correct view to have for this market? What should it be, Jason?
03:58Well, just contrasting two different perspectives. So yes, you have from an engineer's perspective,
04:03this is just a rational response to a binding technical constraint. We need more compute.
04:09The AI we have today, as great as it is, is the dumbest AI we'll ever have. And the capabilities
04:15are going to scale with compute resources. So, you know, the more the merrier. But when viewed from a
04:20purely statistical perspective, the growth, the scale of this CapEx boom is really quite
04:28extraordinary. You know, and you see the graphs and then the hockey stick like graphs. And so from
04:35a time series perspective, it looks like the time series that has entered a multiplicative growth
04:39regime of uncertain duration, which typically points to significant instability ahead. And really what that
04:47means is that it's become very hard from a quantitative risk perspective to measure the range of
04:54outcomes. It's turning into pure guesswork. And that's a practical challenge for investors because
05:01risk premia have not risen commensurately. I'd say that's not being priced in right now. No, no,
05:05that's the issue. It's not, you know, again, two different perspectives. Interesting. But the issue is
05:11is that there has not been the increase in market-wide risk premium. And, you know, if you look at
05:15credit
05:16spreads, if you look at measures of the equity risk premium, they both, of course, trended down
05:21during the period of this buildout as it's intensified. And I think that's the issue today.
05:27Well, the conversation that's happening today is, are we going to have to slow down because we're
05:30fearful of the advance about what it means for society and for humankind? Do you then take
05:35a different approach that it's inevitable we slow down just because of the economics of building this out?
05:41I think a slowdown is more likely than not. Again, these graphs do not persist going vertically
05:47forever. I do think also there's this issue of the question of the frontier models relative to open
05:55weight, the price premium they command. What if some of the frontier models introduce intolerable
06:02cybersecurity risks? Now, potentially at some point in the future, the addressable market,
06:08you know, the revenue base may be somewhat smaller simply because you cannot distribute the models as
06:14broadly as you would like. And so that's an issue. It's really thinking about model capability and
06:21thinking about monetization. They're not one in the same. And I think that's what people are trying to
06:26sort out at the moment. Well, at the moment, it has been an arms race. So that would be kind
06:30of a
06:30different tack for these large language model labs to say, OK, we're going to back off spending on the
06:36most very expensive models because of what you're saying, because of the economics. How does that
06:40triple trickle down throughout the ecosystem of AI? For example, the huge data center build out
06:46we're seeing now, does that also necessarily slow? Well, I think, again, when you look at the growth
06:52in CapEx, it's, of course, led by the leases. So when you look at remaining performance obligations,
06:59order backlogs, right now from the hyperscalers, the neoclouds, it's about $2.5 trillion.
07:05What's interesting about that is if you take the present value of the revenue from those compute
07:11leases and you compare it to the present value of actually having to build and deliver that capacity
07:17at a time when loaded circuit boards are priced at 180% above year-ago levels, there's only economic
07:25value of about $375 billion. So it's only a margin of about 15%. And so I think if you have
07:33frontier
07:34labs that want to slow or there's worries about access to capital markets necessary to make good
07:40on the lease commitments, or if you run into constraints political because you can't build
07:45the data center because of opposition, or just real resource, you can't plug in the GPUs,
07:52you know, that that margin doesn't provide perhaps enough comfort. And you could see a slowing as a
07:57result of that. Jason, what about all the people who are providing this capital that have said,
08:01look, we're not making a bet on AI itself. We're not making a bet on technology. We're betting on
08:05whether these hyperscalers, whether Google, whether Meta, whether they have credit worthiness. And they
08:09look at their credit profiles and say, look at where NVIDIA is at, for example. We're fine. That's not a
08:14concern to us. It's very true. When you look at the project finance, in many cases, there is, of course,
08:20residual value guarantees provided by investment grade sponsors. There's other credit support wraps
08:27on senior tranches. So this is not necessarily the providers of external finance taking a view
08:35on the economics of the project. They're taking a view on the credit quality of the sponsor.
08:40I do worry, though, because this is reminiscent in some ways of 2005 to 2007. And what I mean by
08:48that
08:49is at the time you had market participants who said, I'm not really taking a view
08:54on the quality of this mortgage loan collateral. I don't need to, because this is sponsored by,
09:00say, Citi. And I know that that Citi has the financial capacity to take this onto the balance
09:07sheet if need be. Eventually, liabilities grew to levels. The implied volatility in that mortgage
09:15collateral expanded and that no longer became feasible. We're certainly not the case today
09:21as it relates to these investment grade sponsors. But again, as the liabilities escalate and as perhaps
09:29the implied volatility or the distribution of outcomes with the revenues associated with the data
09:34market centers widens, we could get there at some point in the future.
Comments