00:00Jason Thomas of Carlyle warning the pot of money financing data centers might already be running dry writing the hyperscalers
00:06free cash flow has now been exhausted over 80% of the projected capex over the next few years will
00:12have to be externally financed. Jason joins us now for more. Jason welcome to the program. I always enjoy your
00:17research. You know that. I think it's great that you publish it on the website as well at least some
00:21of it. Just share with us where we are now and how increasingly dependent we might be on capital markets
00:26and also how wide open that window might be for some of those companies.
00:31Well, I think what's interesting is that in the real economy, we have seen signs of crowding out for the
00:36past several years. And that is essentially the AI data centers bidding away real resources from the rest of the
00:42economy. You can think of, of course, grid connections, cooling systems, engineering talent, and then, you know, construction workers. It's
00:51remarkable that since the Fed started raising rates in 2022, real estate development, traditional, commercial, residential, down by about 12%.
01:00But yet construction employment has risen 3.7 times as much as x construction payrolls. But now this is something
01:09that is, of course, of great interest to market participants and market themes. Because in the past, this was all,
01:15for the most part, internally financed. It was the operating cash of the hyperscalers that was paying for all of
01:21this.
01:21Now that pot of money has largely run dry. We saw Alphabet moving into negative free cash flow for the
01:28first quarter since they went public in 2004.
01:31And so on a prospective basis, you have investors, portfolio managers, that have to decide, is AI, is this trade,
01:40is AI data centers going to consume almost the entirety or certainly a large portion of their incremental risk budget
01:48over the next three to five years?
01:50And I think when you see the upward move in yields, some of the uneasiness we see in the markets,
01:56I think this is a genuine question.
01:58And it may not be quite as easy as many suppose when they looked at those spending projections, you know,
02:04five and a half trillion between now and the end of 2030.
02:08Well, Jason, let's get into that question. So to your point, they've crowded out parts of the real economy.
02:13Markets in America, as you know, very deep, very liquid. Do they have the potential to also crowd out capital
02:17markets, too?
02:18And you touched on the maybe the pushback that started to emerge just on the margin.
02:22How vulnerable do you think parts of capital markets are to the issuance we might see, the capital demands of
02:28these companies?
02:30Well, I think what's interesting is that you don't need to be bearish on the prospective returns, expected returns on
02:38data centers, on AI more broadly.
02:40All you have to be is especially conscious of prudential risk concentration limits and how much of your portfolio you
02:48want concentrated to a single risk factor.
02:50Secondly, what we've learned this year with chips, why is the broader semiconductor complex memory chips selling off?
02:59Ironically, it's because they're making too much money.
03:02When you look at Samsung's guiding towards 19-fold year-over-year growth and earnings, companies traditionally with 20%
03:10to 30% operating margins moving into the 50% to 80% range.
03:15When you look at just the BLS data, loaded circuit boards, the GPUs, plus the high bandwidth memory, then the
03:22capacitors, the labor necessary to construct them, that's up 160% year-over-year.
03:28So that shock, that positive profit shock, price shock to that space, has raised a lot of questions about is
03:38there profitability available downstream?
03:40Can the infrastructure layer, the hyperscalers, neoclouds, recover those costs in compute rental rates?
03:47Can then the frontier labs and the models that are renting that compute actually pass on those costs?
03:54And I think that this is much more complicated because people, instead of thinking in terms of an AI trade,
04:00are a lot more focused on the links in the value chain and accounting identities and how excess profitability in
04:07one space actually makes the world much more complicated downstream from that.
04:11You have an interesting thought, Jason, in your piece, where you think that part of the reason why valuations got
04:18so heady earlier this year, particularly in places like memory chips, was because of some of the actions that the
04:24Fed took.
04:25And this was kind of akin to 1998, in your view.
04:28Can you talk about why you think there is this connection between Fed policy last year and the tech build
04:35-out and the tech valuations that we saw earlier this year?
04:39Sure. I do think it's quite analogous to 1998.
04:43In that year, you had, of course, LTCM failed.
04:47You had the Russian debt default.
04:49The economy was doing very well.
04:50Real GDP growth was about 4%.
04:53So this is the economic data.
04:54We're not suggesting that it was time to cut rates, but there was worries about spillovers from financial markets.
05:00The Fed cut rates three times.
05:02The next year, fixed investment in telecom rose 75% over, I think it was the next 18 months, where
05:09from 98 to the peak in early 2000, the NASDAQ tripled.
05:14So I think that this was a lesson that when you cut rates, rate cuts are a very blunt instrument.
05:21You don't get to channel who benefits or how the lower rates are used in the economy.
05:27When you're cutting into a concentrated CapEx boom, you're actually, you're bringing coal to Newcastle.
05:32You're taking what is already probably above trend fixed investment growth capital accumulation and causing it to go vertical.
05:40And that really is quite similar to what we saw in the first six months of 2025, 2026, following the
05:49three rate cuts last year, where compute investment in the United States went vertical.
05:54You know, the growth as a share of GDP was 63% between September of 2025 and what we expect
06:02to be the results through June 30, 2026.
06:06So I think that that is when you're in the midst of something like this, these sorts of CapEx booms
06:12only come around once every 25 years.
06:15This could be something that we haven't seen in roughly 100 years.
06:19Maybe the last time we saw capital accumulation on the scale was electrification in the 1920s.
06:24When you're in that kind of environment, you have to be a bit more conservative about the rate cuts.
06:30And I think that the discussion today, and I certainly don't expect a rate hike tomorrow, but I think the
06:36discussion is not really about hiking rates.
06:38It's about whether they should take back those cuts that happened last year.
06:44I am wondering, though, Jason, just to kind of bring together the point you were making about correlations and looking
06:51for uncorrelated assets.
06:53Can you put this genie back in the bottle or has everything been reduced to an A.I. trade that
06:57if you do get some sort of reversal of some of the high valuations,
07:01you also take away a lot of the strength in the economy that has supported a lot of other themes
07:06as well?
07:09Well, I think that the issue today when it comes to rates is really that, you know, we've been too
07:15focused on technocratic details.
07:18You know, right now people are talking about how there's going to be revisions to PCE.
07:22And, you know, it's that kind of discussion.
07:24But what we've learned over the past year is that if the central bank won't deliver price stability, voters will
07:34take matters into their own hands.
07:37There's been a 27 percent cumulative increase in the price level since 2021.
07:44Many elected, many politicians are, of course, campaigning on an affordability theme.
07:50And that affordability theme in many ways is suggesting that this increase in prices is a market failure.
07:56And they're recommending very aggressive public intervention in the markets.
08:02And so I think that rather than look at revisions and like, oh, you know, don't worry that your grocery
08:07bill is up 30 percent because there's going to be downward revisions to financial services and legal services and the
08:14new PC reports or everything's fine.
08:16Instead of taking that approach, I think that the Fed should more clearly communicate that ultimately they are accountable to
08:23voters.
08:23We spend so much time talking about Fed independence, you know, the independence from political influence on their decision making.
08:30But I think we should probably spend more time thinking in terms of accountability.
08:34And voters have been upset about excess inflation, cumulative increases in the price level very consistently for a long, long
08:42time.
08:42And so I think, again, it's not just this genie back in the bottle.
08:46I don't suspect, you know, I think this has got a momentum of its own and it's going to play
08:50itself out.
08:51I don't think rate hikes would do anything to undermine that.
08:54But the rate hikes would send a very important signal that the Fed is, again, ultimately accountable to voters and
09:02that they take their price stability mandate very seriously.
09:06Jason, do you think that it could potentially also cap longer term yields in a way that it hasn't been
09:11able to do so over the past few months?
09:14Well, I do think the upward move in yields is partly reflects the experience.
09:19When you look at TIPS yields, you know, 2.4 percent at 10 years, almost 3 percent 30 years.
09:26A lot of that, I think, reflects the beating that TIPS investors took in 2022 to 2023, losses of 20,
09:3430 percent in some cases.
09:37TIPS provide a hedge against inflation.
09:39They do not provide a hedge in the increase in real rates necessary to actually stabilize inflation.
09:47And so I think that the real term premium, what we've seen increase over the past few months, is something
09:54that ultimately could come down if you have a signal from the Fed that they're going to be more serious
10:01about their price stability mandate.
10:03And, of course, when you think about where financial markets are tight today, if there's any evidence of that anywhere,
10:10it is, of course, in mortgage finance.
10:12And because we have 96 percent of outstanding mortgages in the United States that are long-term fixed rate, it
10:19is the 10-year that is the key reference rate there.
10:23And so, you know, it could be that rate hikes, or at least the signal from hikes, actually do cause
10:31some of those reference rates, those 10-year yields to come down a bit and actually improve mortgage finance and
10:38home buying affordability.
10:39And I think that's the key.
Comments