00:00There is a wartime level of spending going on today in data centres and AI.
00:07I am a true believer in the staying power of AI.
00:12A lot of people, and I'm the same, look at the internet and say this is a very close parallel to what we saw with the internet.
00:20I think undoubtedly when you see this level of growth and this level of spending,
00:24just common sense tells you there's got to be speed bumps along the road.
00:29No doubt about it.
00:30And so as an infrastructure investor, what we're trying to do is give a somewhat lower risk way to play that trend.
00:38And so we're always investing behind long term rental contracts with the big hyperscalar tech companies.
00:45So from that sense, I'm rather relaxed about the way we're playing.
00:48I got to say, because I hear this frequently of folks saying, we're not taking AI risk.
00:55We're taking meta risk.
00:57We're taking Oracle credit risk.
00:59Mike, what happens if that changes?
01:02Oracle CDS spreads, for example, have started to widen recently.
01:07They've doubled in the past couple of months, of course, to a level that's not crazy.
01:10But the market is starting to price in more risk for some of these hyperscalars.
01:14Is that a problem even within itself of taking risk on these companies that are so concentrated and spending so much money?
01:23I'm not too worried about taking credit risk on the biggest companies on the planet, matter, Google, Amazon, et cetera.
01:35I would be there's obviously a different way to analyze the more startup LLM companies that arguably is a different type of risk profile.
01:44And so I'm probably more relaxed just from a pure credit perspective about taking the trillion dollar plus big tech companies.
01:53There is, though, so much funding needed to continue to provide all the CapEx for all of this AI stuff.
02:02It's like in the hundreds and hundreds of billions of dollars.
02:06And that's what everyone is watching.
02:07Everyone is saying, hey, is the tap at some point going to be turned off from a capital markets perspective?
02:13So that's what I probably would keep my eye on.
02:16Well, the root of it is the data, right?
02:18I mean, I got hired at this company in 1999, and we were all so excited with the fiber cable getting laid out all over the place,
02:28paying as much as we possibly could, right, because Internet was going to be the next big thing.
02:31And Internet was the next big thing.
02:33And yet we still laid way too much fiber, hundreds of thousands of miles of dark fiber for years before that demand really showed up.
02:44Could the same thing happen here?
02:48There are absolutely parallels between the Internet era and the AI era.
02:54And I do think it's yet to be proven which business models are going to work in the world of AI.
03:00There's no doubt in my mind that there is enough.
03:03If you just look at the cost savings from a labor standpoint, for example,
03:07there's no doubt in my mind that all this spending over time will earn a return just from labor savings alone.
03:15Now, whether or not the folks who are spending that money are the ones who get the savings or get the earnings from those savings,
03:21I think time will tell.
03:23I will make one big distinction, though, between the Internet boom and the AI boom.
03:30There's a whole bunch of startup Internet companies that had no internal funding ability.
03:36All of their funding was external.
03:38You've got very much the opposite story with a lot of the AI companies.
03:42You've got the biggest companies on the planet with the biggest cash flows on the planet that are central to all the spending on AI.
03:49So I do think it's more robust than the Internet.
03:53I mean, they're going to keep buying picks and shovels, these hyperscalers, but are they going to strike gold is the question, right?
03:59Or does it take them 10 years to strike gold, which is what we saw happen from 2000 until we didn't really recover that spend until like 2013?
04:09Well, I think if you're Google or Amazon or Oracle or one of these big cloud companies, you're going to do very well in your cloud business just from basic cloud.
04:20But AI is going to put fuel on the fire there.
04:22So I think they're already printing money from an AI perspective.
04:26But your question is very pertinent.
04:27And what happens in all of these big gold rushes is initially it's all about growth.
04:32Whoever's growing the quickest gets the stock price shooting up.
04:36Eventually, people look at return on capital.
04:39That's where things will get to ultimately.
04:41And it's just a question of is that next month or in two years or in 10 years time?
04:47This goes back to what you said before.
04:48Maybe the thing to keep your eye on is whether capital markets stay open for these hyperscalers that are tapping them to fund.
04:54What are the potential things that would cause capital markets, be it private players who are financing this, be it broadly syndicate markets, that would cause them to step back and maybe not as willingly price some of this debt?
05:09I think the big one is simply market sentiment around am I happy to keep funding growth with a future promise of returns or do I actually want to see returns?
05:21Now, if you look at a very good parallel is, say, the fracking boom in the US where this is all the oil and gas drilling, where for years and years it was all about can I drill more oil, can I drill more gas?
05:32And then ultimately investors said, hey, where are the dividends for me?
05:36So when I turn to the tech world, I do think tech investors are willing to bet more on the come.
05:42In other words, I think tech investors, just by their very nature, are willing to invest on a longer time frame.
05:50But again, it's that sentiment change from growth to return on capital.
05:55Maybe the other thing I would throw in there is that if the AI tech hits a speed bump, like if it doesn't deliver the sort of improvements that optimistic people are saying it will deliver in short time frames,
06:09that as well may cause a bump in the road.
06:12I do think long term it's going to be, you know, it's going to reflect the true belief that so many of us have.
06:19But it's so hard to tell the little cyclical path it'll take to get there.
06:24Mike, power bottlenecks were pointed out nice and early and investors have really piled in to those plays.
06:31Where is the next investable choke point in AI?
06:36Is it like transmission, interconnects, affirming, gas storage?
06:42It's all of that.
06:43So I believe today that the best way to play the power constraint in AI in scale is through natural gas in the U.S.
06:53simply because you can invest in that today.
06:55It's very scalable.
06:56All of the power that we need to power AI, an awful lot of that is going to come from natural gas.
07:04There's really four big bottlenecks from a power perspective.
07:08Firstly, it takes a long time to get the generation units.
07:10If you want to build a power generation unit today, you've got to go and buy the actual turbines from Siemens and GE.
07:17You're lucky to get one in 2028 or 2029.
07:20Long waiting list.
07:21Firstly, secondly, so many projects want to get on the grid because not only do you have all the AI related power projects, but all of the renewable projects as well.
07:29It's three to five years to get on the grid today.
07:31So all of the operators of the grids around the U.S., they're looking at ways to speed up your ability to get on the grid.
07:37That's a big one.
07:40The opposition from the administration towards renewables is actually a constraint on AI power at the moment.
07:48The quickest way to get power on the grid is solar.
07:50And so you're seeing two policy aims run into each other where you've got really, really strong policy support for the growth of AI.
07:58We want to make sure we're staying ahead of the rest of the world.
08:00But on the other hand, you've got policy, what's the right word, constraints on renewables in the U.S.
08:08I do think that there's maybe some grand bargain to be had between Republicans and Democrats around that, which brings me to the last constraint, which is the transmission grid.
08:17It's very difficult to get permitted to build new transmission grid.
08:21And we've got very close a couple of times federally to streamlining the permitting process.
08:26And I do think there's a bargain to be had where the Republicans say, hey, streamline this process for us.
08:31And the Democrats say, fine, but ease up a little bit on renewables.
08:35So that's the way I think about the power constraints today.
Be the first to comment