What Happens If Data Center Construction Stops?

[00:00:00] Speaker 1: There is some obvious concern this week over, at the very least, slowing down of data center construction or perhaps even stopping data center construction in some states. We're going to break down what you need to know as an investor in this video. What we're not going to talk about, [00:00:17] Jacob Goldstein: politics. Hopefully that's obvious. We're a pretty much no hype type of channel. If you're new here, well, there you go. Now you know. So anything that stirs up emotional decision making, we're not going to do that. So what we will talk about here is this trend that was actually already in place before midterm elections even happened. All the way back in March, over on Semiconductor Insider, we had special guest Jacob Goldstein. A portion of that interview was released on April 2nd. We talked about the Luddites. A couple hundred years ago, they were pretty upset about the pace of technology and its disruption of their life. And so they started smashing things up. And we certainly didn't say that was going to happen with data centers, but people kind of don't want these big, ugly data centers built next to their home. This is not a new trend, is the point. So this is just really a continuation of the trend that was already in place. And so that we can actually talk about and discuss if this trend were to maybe accelerate, as Casey said, new construction starts to slow down. Does that mean it's the end of the so-called AI trade and the AI data center growth trend that we've been enjoying the last few years? Simple answer, no, it's not. Let's break it down. [00:01:40] Speaker 1: So make sure you check out that episode with Jacob Goldstein. We'll link it in the video description, hit the subscribe button and check out chipstockinvestor.com where we have our semiconductor insider access, where you get all of our research tools, all of our research and a weekly live Q&A event. The first thing that we need to differentiate here is new data center construction halting versus maybe existing data centers. The work that can be done in those data centers, their shell is already perhaps built or even further, the data center is already in operation. Those are two very different problems. Yeah. A lot of the concern is about new data center [00:02:24] Jacob Goldstein: construction halting. And companies can still obviously work with a data center that they actually already have built and most certainly one that's already up and running. There are still some limitations on, for example, getting power into that data center, but that's a very, very different type of problem than, hey, we have this land or we would like to acquire this land and develop it. That's more the risk. And so like we've been discussing all along, should something like this come up, who benefits? Well, the companies that have lots of computing power already up and running, especially the hyperscalers. And lo and behold, while a lot of the AI trade got clobbered immediately following midterms, really just kind of continuing the volatility from October and the last four months, those stocks actually overall did pretty well. Who knows what happens by the time this video actually publishes. It's pretty normal to experience some volatility up and down immediately following political events. But that statement is probably going to hold true. If you have existing data center assets already and everybody else can't build and catch up to you, suddenly you have an advantage. And it would appear early on the market has woken up and decided it agrees with that. [00:03:48] Speaker 1: So what's one thing that you can do in your existing data center? Well, it's increase the compute power. On the left side of this chart, we have what we call the legacy hall, the legacy data center and how much compute and energy is needed for that compute, all the way up to the next gen racks. Think of the Vera Rubin by Nvidia, those racks. They require a lot more power, but obviously put out a ton more compute. So that's one thing that an existing data center can do. They can add to the compute load that they already have if they're already hooked up to the grid. And this has been one of Nvidia's big [00:04:29] Jacob Goldstein: arguments all along. Even if you can't increase the amount of power going into that data center, well, you can upgrade older racks and maybe you're still ultimately using the same amount of electricity to power the whole building. But because you've upgraded to a more powerful system, you're cranking out a lot more compute power. And as Jensen famously said earlier this year at GTC, in the AI era, compute equals revenue. So there you go. There's one thing an existing data center could do if they can't build new. Well, they can upgrade the old. [00:05:07] Speaker 1: The second thing that a data center can do is upgrade the power systems and the cooling systems to the data center itself, as well as upgrades to the building, the structure itself. [00:05:19] Jacob Goldstein: We started discussing this well over a year ago. We'll start on the left. Power. Nvidia has been working on this high voltage data center architecture for quite some time. Now, this doesn't necessarily mean a company needs to get more power into the building. They can also work on this with the same existing amount of power or contracted power if they have a utility provider. Or it could be on-site power either with a natural gas generator or it could be something from like Bloom Energy with a fuel cell. Either way, if you want to upgrade to the next-gen servers, get more computing power for the same amount of power or also more power plus an upgrade to your internal systems, your racks, you would have to start here on the left. Bigger switch gear, overhead busway for running the cables, and then also equipment for higher voltage power at the rack and the servers within them themselves. [00:06:19] Speaker 1: The second thing that you can do, as I mentioned, is cooling. Right now, air cooling is going to be maxed out very soon with these new systems, these new racks. It needs a different type of cooling system, like liquid cooling, which could be at the chip or at the server. Coolant distribution units are needed, or those are the CDUs, and a liquid loop out to the chillers so that all that heat can run out of the data center and outside of the facility themselves. Companies like Vertiv or Flex are a couple of companies that provide these chilling or cooling systems. And then finally, the building itself, [00:07:01] Jacob Goldstein: would obviously require some permitting. No doubt. No doubt to expand the building. But it is another option. If the company has run out of space, they can add on to it in some way. Pull in a construction team, and now they have some more space to add on to the existing equipment or build a new section of the data center. Reinforce the strength of the floors. Have some more room for the power and cooling. The first two boxes there on the left. Casey's handy-dandy schematic of what we were just looking at. Everyone knows I love schematics. [00:07:39] Speaker 1: But here is a schematic of what this looks like. We just have it broken up between the green, which is the power, [00:07:45] Jacob Goldstein: cyan, which is the cooling factor. Just let's call it blue. No, cyan. It's the blue part of the schematic. [00:07:52] Speaker 1: It's cyan. And then at yellow, the building. This should just help you visualize what type of upgrades [00:07:58] Jacob Goldstein: can happen in an existing data center. And we should throw more one thing in the mix here. Besides these companies being able to work with and upgrade their existing data centers, there's also this whole part of the argument where if a certain part of a state says, "No more new construction," or, "You know what? We're going to go further and tell you, you can't do anything with the existing one you have either." Well, there might be another place within the United States that says, "Hey, we'll work with you on this." So there's also that too. This is not an end of the story. We do think this does perhaps provide a sign that we are nearing the end of the capex to revenue cycle. A lot of these companies have been shelling out a tremendous amount of capex, and it has caused the market a lot of worry and concern. Perhaps this is just the trigger for the hyperscalers in particular to start to slow down or allow their cloud and AI revenue to start to grow at a faster pace than the capex itself. That would be a fantastic sign. And again, for quite some time now, over two years, we've been talking about this being a soft hedge against the semiconductor companies. And that actually suddenly worked out pretty well, at least day one or two after this political result. I wish there was someone here to tell us these things. A couple more slides for you. Let's talk about two companies that had fantastic earnings in Q3, 2026. We're not going to dig into the numbers because that's just too boring. Not really adding much intellectual discussion or critical thinking because you've probably seen about a thousand social media accounts or media articles providing a review of management's earnings review. I don't think we need to do that. Let's just talk about a couple of points. First, Arista Networks. [00:09:56] Speaker 1: We've done a little bit of a rework on this three parts of a computing system from our old design, but this illustrates the three parts of the system, the computing, network, and storage systems. And Arista now plays in all three. While Arista is a networking company, they're now building systems that work with the other parts of the computer, the computing, as well as the storage system. Ethernet fabrics is what meshes all of that together. And so the first few years of this AI [00:10:32] Jacob Goldstein: data center build out, GPUs and accelerated computing were the focus and it remains the focus. But Arista said on their earnings call, they're now expected to be working with over 10 AI accelerator companies on the compute side of this equation, various storage solutions as well. And you can see the early signs here of what happened in the past, in the early advent of the cloud and turning data centers into these really powerful assets for software as a service. They're abstracting away the need for companies to be preoccupied and concerned with what's actually happening on an individual component basis, individual server, even individual rack basis within their data center. And Arista can just architect the whole thing for them. And the customer can use Arista's software and services to manage the whole thing. You get cyber security and analytics in the mix. They have a new MCP product that allows you to interact with your network in natural language, and it'll go out and just find out what's happening within your network. If you're not a tech business and you have this really powerful tech asset, you don't know what's going on. Arista's MCP will go out and just tell you in also natural language and then help remedy a fix. Pretty incredible solution. We had that big two-part article series two years ago, exactly. Breaking down the brewing battle between NVIDIA and Arista. We contended that Arista is a very different type of asset, very different type of business than NVIDIA, even though NVIDIA was encroaching on Arista's turf. And lo and behold, two years later, far from being disrupted, Arista has thrived and is working on a lot of new stuff that actually brings new competition, potential disruption, to NVIDIA itself, rather than just the other way around, NVIDIA potentially disrupting Arista. So we continue to hold both in our portfolio and think there's a really good case to be made for continuing to hold both. There's not redundancy here or so much overlap between the two companies that you have to choose just one. They're both great companies and doing a lot to advance the AI data center and AI network technology and capabilities for their customers. [00:12:59] Speaker 1: : And one final company to round out our discussion today is Monolithic power systems. Monolithic makes power chips. Actually, Monolithic has a best-in-class catalog of not only power chips, but also systems that can be placed absolutely anywhere that needs power. This can be, of course, next to the GPU, in the rack, in the devices that give the power to the rack, robotics, industrial, automotive, literally everything. Monolithic power has a power system or power chip that can give that device power. [00:13:33] Jacob Goldstein: : Really incredible story. They continue to just go from strength to strength and a best-in-class mix of not just revenue growth. As a fabulous chip designer, you would expect that, the ability for them to move very quickly, reallocate their engineering teams to capture new design wins, but also best-in-class profitability. That's oftentimes a sour point for high-growth tech companies like this, but Monolithic is actually profitable on all counts and on a per-share basis over time. Though they've been making some cash investments this year to expand with their manufacturing partners, still very much delivering on this front, not just growth, but fantastic profitability for that growth and returning cash to shareholders with share repurchases. This is one that after talking about it now for six years, seven years, still kind of amazed that this one kind of falls by the wayside and just gets overlooked. We're actually totally cool with that. We're fine with that. We're no hype people. We like no hype businesses and no hype stocks [00:14:41] Speaker 1: too. Just deliver the results. Make sure you hit that subscribe button. And if you want to take a closer look at these slides or all of our research, make sure you go over and check out Semi Insider on chipstockinvestor.com. See y'all soon. Take care.