The Global Supply Chain for Data Centers is More Complicated Than You Think

[00:00:00] Speaker 1: People just assume that you build a building, you put in a bunch of equipment, and the blinky lights show up, and it just happens. There are so many dependencies when deploying any of these things. The most obvious one is you are waiting for power to be delivered to your space, and what you're waiting on is the supply chain of those. But the same is true not just for, let's say, the UPS gear or the cabinets, but the components that go into building those very UPSs and cabinets and transformers and the cooling gear and all that. That's before you even get to the supply chain associated with the compute workloads and all that. [00:00:47] David Snow: Hello, and welcome to Cool Vector Hot Takes. I'm David Snow, your host. And today we're joined by John Wilson, the CFO of Sebi Data Centers. Philbert Shi, founder of Structure Research. Nabil Mahmood, from the Nomad Futurist, as well as a Cool Vector editorial advisor. And another Cool Vector editorial advisor, Philip Koblenz, from Critical Ventures, also a member of Nomad Futurist. So everyone, welcome to Cool Vector today, and I'm looking forward to delving into a very important topic, which is the supply chain of the global data center business. We're definitely not going to be predicting any political moves. What we are going to do is learn more about what the true nature is, the global nature of the data center supply chain, and talk about what might happen given any number of political directions that this world goes. Let's start with a question for John Wilson from Sebi Data Centers. John, in addition to being the CFO, you were also in charge of procurement. And so you have a unique view on what it takes to get a data center project going here in the U.S. What do you think investors would be surprised or interested to learn about the supply chain for data centers? [00:01:58] John Wilson: One of the things that I think is often overlooked when people think about the data center supply chain is not really just the physical data center itself. It's not really just the servers and racks associated with it. But more and more, when we think about the supply chain, we have to think about the energy and electricity production and the impact that that has on our ability to service the demand that we're seeing in the data center space. So we have to think about not just how do we secure the supply for the labor to build our data centers. Not only do we have to think about servers and the racks that our customers are going to bring into our data centers, but at the very front end of that, we have to do more and more work with the utilities to think about their supply of electricity into our data centers and the supply chain that they're navigating to bring in transformers and other electrical equipment to be able to deliver that power to us. It becomes incredibly complex when you think about all the different moving parts that we have to navigate. Energy production in the United States has been almost flat the last 25 years or so. I think it's grown 13% since 2000. As we shift into an era where more and more items are electrified, including data centers, we're asking these utilities to think about their business in a way that they haven't had to think about it in at least 25 years. So that's complicating the supply chain and the overall story that we have as we think about building and constructing the infrastructure for the AI and digital economies. I would not have guessed the answer to the question [00:03:33] Speaker 1: is the complexity is not just the supply chain, but the supply chain of supply chains. It makes total sense. People just assume that you build a building, you put in a bunch of equipment and the blinky lights show up and it just happens. There are so many dependencies when deploying any of these things. The most obvious one is you are waiting for power to be delivered to your space and what you're waiting on is the supply chain of those power companies. But the same is true, not just for, let's say, the UPS gear or the cabinets, but the components that go into building those very UPSs and cabinets and transformers and the cooling gear and all that. That's before you even get to the supply chain associated with the compute workloads and all that. The fundamental point that all of these industries are interrelated, all of them are dependent on factors that, regardless of what our desires are, are completely outside of our control. What we're trying to do as an industry is under-promise and over-deliver to the extent possible, setting expectations amongst investors, end users, and everyone in between so that you don't have a construction crew that is waiting at a site that is dependent upon stuff that isn't there. And you end up going over the top on cost overruns by virtue of not effectively communicating expectations up and down the value chain. [00:04:54] David Snow: It's certainly the case that the data center industry is incredibly collaborative. You need so many different forms of expertise and input to get one of these things built. Philbert Xi from Structure Research, what can you tell us about the supply chain feeding into data centers that you think investors would be interested or surprised to learn? [00:05:13] Speaker 4: They may be very surprised to see how complicated it is. If there is one bit of disruption, as we saw during COVID, there's almost like a cascading effect across different parts of the supply chain. It's almost like whack-a-mole, right? You might have two of the things figured out and then if the customer doesn't have chips, doesn't have servers, doesn't have cables, they're not going to be able to set up their infrastructure in the data center. Just one little disruption and these things are not going to be [00:05:36] Speaker 1: done on time at cost. I think the one thing that investors will always be surprised by is that money doesn't solve every problem. It turns out that the money is the easy part, especially when you have the amount of demand that we're seeing in our industry. It is amazingly nuanced, and it's not really just about the investment and the dollar, but about understanding what the realities on the [00:05:57] John Wilson: ground are. But as Phil pointed out, this isn't stuff that, you know, even if you're throwing multiple billions of dollars at it, you can solve in an instant. It takes time to recreate these supply chains. It takes time to reshore manufacturing if that's really going to be the long-term solution. These aren't problems that are easily solved in a moment. It takes careful planning, it takes work to find alternatives, and it takes patience, unfortunately, in many cases. [00:06:22] Speaker 5: The supply chain really is very much global in nature. It's very intertwined. Like one component misses. You can't stop the entire production or you can't really build a data center. So the challenge really is the lack of understanding of the complexity of product development. I mean, products might be designed or engineered in America, but they might be assembled in South Korea or China, and then parts might be coming out of some other parts of the world. What we're trying to accomplish is to demystify that process. It's a process that's going to take time. I was hoping that by COVID-19 or the end thereof, we would have actually learned the complexity of this process, but we are back to it again with the tariffs issue. [00:07:02] David Snow: Sounds like COVID-19 was a dress rehearsal for what we're facing now. John at SABI data centers, I know that the answer to this is there are many parts, but can you think of any particular pieces of equipment that are made outside of the U.S. that are going to be prime disruptors of the process if there becomes some sort of impediment to these parts making it to the U.S.? [00:07:23] John Wilson: There are plenty of components that are manufactured in the U.S. I mean, generators are a good example of that, but there's almost nothing that doesn't have some subcomponent or material that isn't sourced globally. What's most impactful, it's the UPS systems, it's the EDCs that we generally bring in from outside of the U.S. So those are the ones that are causing us the most concern in this particular moment in time. If I think up and down the supply chain, it's the transformers for the electrical supply, it's the GPUs and the CPUs for folks that are populating the servers in our data centers as well, that I would put right at the top of that criticality list. [00:08:02] Speaker 1: How much of these things can even be solved by finding your new suppliers? I imagine that every supplier has the same level of constraints and maybe if you find an incumbent to the U.S. market, maybe they don't have the backlog that usual suspects, the Vertives and the Schneiders of the world have, but is there even a benefit to trying to work outside of the usual suspects from a procurement perspective? [00:08:27] John Wilson: There are things that you can do around the edges to optimize that supply chain. It is very difficult to introduce new suppliers into the space. We have done that on fairly rare occasions. It's a hard conversation for us to go and talk to a hyperscaler and say, "Don't worry, your whole system is being backed up by this brand new producer of this piece of critical equipment that's never been tested in a live environment before. How would you feel about being a guinea pig on that?" That's not a winning conversation to have until you really do want to focus on those folks who are tried and true and reliable in an industry that's all about reliability and uptime. It's all about planning, working closely with our partnerships and making those improvements where you can, but they're improvements at the edges rather than fundamental shifts and turnarounds of those supply chain constraints that we're dealing with. [00:09:22] Speaker 5: As a sector, that mindset needs to change. I mean, the technologies are going to be tested and read it out. I mean, there are processes, accreditation certifications that these suppliers do go through, will go through. It's a matter of mind change, right? And also, as data centers end up becoming commodities, which is going to happen within the next decade, the resiliency of uptime shifts. That load shifts from the physical layer to the application stack. So as long as I can get you your data in a timely fashion and I keep the data integrity, it's irrelevant who the supplier was. And a simple example really is, do you guys know who manufactures your glow plugs in your vehicle? You don't. Who manufactures the braid pads? Who manufactures the clippers? The guy that's driving a Formula One car does know that. So it's a change in audience is changing behavior. And as a sector, I believe where we're at and the rate that we're [00:10:16] Speaker 4: drawing that that mind shift needs to change. Thinking about some of the more specialized vendors that data center operators use. So for example, Align Data Centers uses a low carbon concrete supplier that's based in Canada. That is a key component to them in their design, in their building, the carbon footprint that they believe they can deliver to Hyperscale and other types of customers who have these kind of stated targets. So that's just kind of one example of how far reaching this could go. If you don't have a certain piece of equipment, how does that impact your design? How does it impact the engineering and the plan that you've done? Are you at the beginning or are you halfway through that? The time to market your delivery timeline is core to your value proposition. It's make or break in many cases when you're dealing with some of these very large customers, particularly in the Hyperscale community. [00:11:06] Speaker 1: One question that I think is super interesting is whether the supply chain issues, because of the tariff situation in the U.S. particularly, is it going to stifle U.S. development, maybe enhance development in other international markets? I think Sebi is more tied to the U.S. markets, right? You don't have any international developments at this point. Is that right, John? [00:11:28] John Wilson: We are a U.S.-based data center provider. Near-term, we have not seen any slowdown in demand at all. I think there was a brief pause immediately after the tariffs were announced. Candidly, we've not seen any slackening in that demand. There are three tiers that we look at and think about as a business. First is enterprise platform and what's happening on enterprise scale-out. That's growing more slowly than the other vectors, but it is still continuing to grow. Public cloud build-out is still happening, you know, double-digit CAGR growth in that space as well, which I think a lot of people forget about, right? We talked so much about AI, which certainly is a huge tailwind to our industry and a big driver for a lot of the demand. But there's also a lot of demand on continued build-out of the public cloud. And then, of course, lastly, is that AI build-out and the very rapid acceleration on that. And in particular, on the AI side, a real desire for the companies who are pursuing that to continue full speed ahead. The stakes of being second place or third place in the AI build-out are pretty significant, and so people have proceeded without really much applause. John mentioned from the enterprise, [00:12:42] Speaker 1: through the hyperscalers, through these kind of AI campuses or factories or whatever you want to refer to them as. All of a sudden, you have the US, which essentially puts a huge barrier to the imports of a lot of this equipment. Does that then accelerate some of the developments outside that might be able to take advantage of the fact that the US has put up this barrier to some of those imports, and then that excess capacity of equipment can then be reallocated to these international markets? [00:13:11] Speaker 6: Yeah, I mean, that's a massive question. Just yes or no answer, Robert. I don't think there's really any yes or no here. You may see the hyperscale clients that are busy [00:13:21] Speaker 4: expanding around the world make decisions based on where the supply chain impact maybe is lessened. It's much more nationalized these days, right? It was once upon a time where you could serve infrastructure from one country to another regionally. The same thing happened here in Canada. Hyperscale was late to this country because a lot of small to mid-sized organizations use US-based cloud services. Once the kind of the issue of data location and privacy, and of course, performance tied into physical location and proximity, you saw that cloud market and the hyperscale market start to take off, and hence the data center market start to really grow in Canada. But because, sorry not to get too off topic, but because of that focus on data sovereignty, I think you'll see that both the operators and the hyperscalers are going to be making different kinds of decisions. Some markets are going to be quite difficult if they just can't get the building blocks. [00:14:07] Speaker 5: I think where we started this conversation was the limitation of energy as it stands within the United States. That's the biggest barrier that we've got. As it entails the products, they can be sourced in other parts of the world. Energy in itself is actually a major inhibitor, whereby that's going to drive a significant amount of development and capital in other markets. And those markets would be Africa, Middle East, and Australasia. And we are seeing a significant amount of investment, particularly from the Saudi Investment Fund and others, developing in those regions. So I do see in the very near future, a lot of investment as it entails the data centers and data center buildouts moving into those regions and those parts of the world. [00:14:48] David Snow: Paul Jay: Philip Koblenz earlier mentioned that money doesn't solve everything. And yet when there are tariffs and when there are parts in short supply, one way to get things done faster is to pay more. Again, there's limits to it. But in these data center development projects, certainly time is important if an extended timeline eats into the returns of the investors. But how about if parts and equipment just becomes more expensive? How would that change a project? Would certain projects get canceled? Would you just have to call up the customer and say, this is going to cost more? [00:15:20] Speaker 1: Philip Koblenz: Building a bespoke data center on behalf of a net least user, like a hyperscaler, where they are the primary user, where a lot of the margin, as it were, is wrapped up in the design, being funded by a very kind of narrow definition of what that one user needs to achieve. So I imagine you work hand in hand with them versus say, you know, a multi-tenant data center site that's going to sell to a broad range of enterprises that can probably absorb cost increases based on longer term visibility into, you know, what their margins are going forward. [00:15:56] John Wilson: Philip Koblenz: I have a couple of advantages that I think help us not to make this too much of a commercial, but we do operate in multi-tenant data centers. So that can help us kind of think about how we absorb some of those costs differently. We also have an integrated design and construction arm as part of our operations. And so we can work with them, think through the value engineering piece of it along the way so that we can get those costs down to what we say are an acceptable level. But at the end of the day, it is for us, just as it is for our customers, a question of what is the return on investment? What is the yield that we're getting off of this project? And if those numbers don't pencil where they, to where they need to be, then the project doesn't work. We need to either find a different strategy or we need to cancel the project. You know, we're not going to do projects that lose the company money. [00:16:45] Speaker 4: David O' We're in the middle of the first quarter earning season and Meta put out a really interesting adjustment to their CapEx and then some commentary. So they raised their CapEx, but they raised, they widened the range. So I think it was like a $4 billion range now, about seven or eight. And the commentary from management was that this is being done. We have a widened CapEx band to account for cost increases and overruns based on unpredictability related to global supply chains and global trade issues. I think that's already an early sign of where things are headed, at least when it comes to some of the bigger, bigger projects that are out there. [00:17:19] Speaker 1: David O' Once prices go up, do they ever really come back down? Inevitably, the one thing that we know is prices will always go up. Maybe it's the pace at which they go up that might shift a little bit, but the damage, depending on what side of the table you're sitting on, is already done. There's just no scenario that I can foresee where prices come back down or people don't utilize either tariffs or supply chains as a justification to continue to increase costs across the board from the operations of a data center all the way through construction and deployment. [00:17:53] John Wilson: David O' Certainly in the data center space, in the recent years, we've seen not a lot of elasticity on pricing due to the strength of the demand. That being said, a couple of years ago, it looked and felt a little bit differently, and prices were much lower two, two and a half years ago than they are today. I guess I don't fully subscribe to once prices are elevated, they will stay elevated forever. I do think that it is much more difficult to decrease prices than it is to increase prices. The upward pressure that we're seeing from the supply chain constraints, now the the uncertainty around the tariffs, all of that does add to persistent inflation in the data center market [00:18:36] David Snow: over the long haul. David O' Can any of you, maybe starting with John, think of a scenario that could happen, like a certain key item of equipment or perhaps something on the power provider side that would blow up a project that would force the leaders of the project to say, "Guys, we're going to have to [00:18:53] John Wilson: mothball this." The most obvious answer to that is the ability to deliver supply. I think we've all seen the news of some of the hyperscalers that have pulled back their demand forecast, allegedly canceling leases, which is a bit of a misnomer, but I think the reality is what's happened there is associated with the delivery of power and that not really coming to fruition or coming to fruition so far out in the future that it no longer becomes worthwhile to pursue that project. So I think that's at the top of the list. And then the other one would be the ability to provide a redundant power, you know, particularly around the generators and not having those in place would require a very different conversation with our tenants for whatever reason. There was a break in the supply chain and we weren't able to bring generators to our data centers. That'd be a very difficult conversation, probably would a mothballed project in most cases. Part of the reason we spend as much time as we do developing those relationships and monitoring the supply chain so we can be as predictive as we can on the delivery [00:19:58] Speaker 4: timing. I mean, I've spent a lot of time thinking about the GPU supply. If there were a situation where that gets severely disrupted, a lot of the infrastructure is being built to house and power these chips. The projections are based on these things being available. If you think about, again, where CPUs, GPUs, where a lot of this production takes place in parts of Asia that are very potentially susceptible to some very, shall we say, dangerous geopolitical situations that would have a cascading effect from that part of the world to this one. So what we're talking about is if China invades Taiwan, [00:20:30] Speaker 5: is it basically what we're talking about? Well, don't get with people on any idea. [00:20:34] Speaker 4: But even, yeah, let's assume that the costs skyrocket because of that. NVIDIA is, you know, starting to move some of their production into the US. How long is that going to take for that [00:20:43] Speaker 1: to kind of get up and running? It takes, you know, a decade to create a factory that is capable of, you know, producing those kinds of chips. And that's without even like understanding, you know, the rare earth minerals and resources that are needed to create that stuff, which, you know, you cannot create a factory in the US that will create that stuff. So if there's, you know, less supply, [00:21:04] Speaker 4: what happens to the costs. And this is even before kind of any disastrous geopolitical situation. I mean, this could quickly kind of snowball to a situation that isn't, is not so great. [00:21:13] Speaker 1: And it's difficult for someone that is building data centers to, to really understand the nuances of the chip market and, and what people are doing with those chips and application development. And not to mention, as Philbert mentioned, the geopolitical strife associated with where all of these materials are harvested. It's understanding that entire ecosystem and, and how dependent on one [00:21:36] Speaker 5: another they all are. I would just say that it's not just the chip in itself, it's everything else. I mean, there have been holdups with EC fast, they've been hold up with embedded DC adapters, right? Every single element is equally important. Positive note, the US is actually expanding its semicolon in chip manufacturing. But then again, we're probably at least five to seven years out before it's, ready for mass production. That addresses one piece of the equation, but there's a lot more that's equally important and we should pay attention to all of it equally. With the tariff situation, what's going on with geopolitical concern, we've got to step up a bit at 145% tariff increase. We're going to start pricing ourselves out of any new data center design build. [00:22:22] David Snow: Jon Wilson from SEBI data centers. Is there any particular indicator or a type of statistic that you're following? I know that SEBI has an in-house engineering and development group. Are you able to look at that dashboard and what are you following most closely? [00:22:35] John Wilson: Jon Wilson: Definitely a number of indicators we look at to kind of assess the health of the business and the health of our supply chain. Both, you know, we, of course, look at GDP numbers, you look at consumer spending, just to get general sense of overall performance of the economy. You know, we look at corporate earnings and what are companies announcing with regard to CapEx and particularly cap-related to data center spend. Within our own business, we're monitoring very carefully our OpEx per megawatt and our dollars per kilowatt for construction of data centers. The challenge with a lot of those is any of these indicators were rearward facing. What we're facing with this latest economic shock with the tariffs, we really haven't felt most of the impact of that. Most of the companies that were announcing earnings, most of the economic indicators were all Q1 indicators, right? So they haven't really taken into account what's been happening since April 2nd. And even April 2nd is, I'd say, a false starting point for this analysis since there was uncertainty. I mean, there's a 90-day pause for some countries, but not for others. It only started the clock ticking when shipments were leaving the ports of origin. And so I think there's still a lot to be figured out as we look at not just the, I'll say, the rearward indicators, but also what's happening in front of us. So give you kind of a perfect example that I happened to be flying back into Seattle last night and the plane flew over the ports probably the first time in a really long time. The port was empty. There were no ships coming in. That's an impact of goods coming in from Asia. The demand for that really sprung down as a result of these tariffs. So how do we figure out, you know, what are those indicators we're looking at and how do we ensure that they're giving us the right projection into the future? Candidly, it's a challenge. There are a number of things we're looking at. I don't know that we [00:24:34] David Snow: nor anybody else is a perfect crystal ball though. What did you learn from the disruption to the supply chain during COVID-19 that you think is applicable to a prediction about what's going to happen given perhaps a tariff disruption to the supply chain? So COVID-19 was a great example. I think one of the things [00:24:51] Speaker 5: that we learned is that we need to build outside of the country. There are going to be a lot of buildouts in other parts of the world like we talked about earlier. One of the key lessons that I learned and I think I'm a big advocate of is the fact that we need to bring business, we need to bring opportunities, we need to bring manufacturing back into the United States. So semiconductors is a good example and that applies to any other nation as well. It's not just that we are sitting in the United States. We need to source locally. We need to build locally. We need to engineer locally. So that's the biggest lesson learned. We need to take care of our household first before we worry about the rest of the world. [00:25:25] Speaker 1: And the data center industry has proven itself to be incredibly resilient. What COVID showed us is that, you know, there is only increasing needs for digital infrastructure. So I don't think there's really a concern about overarching demand, certainly over the long term. Whether it gets consumed at the speed with which, you know, the current AI hype cycle suggests it'll be consumed is almost irrelevant. It's not like the disruption of, say, the turn of the century in 2000 when, you know, people were concerned about whether digital infrastructure, well, you know, the internet was a fad or not. It's here to stay. Those companies that are able to be more flexible in terms of how they respond to shifts in both, you know, workload needs, you know, regional requirements, disruptions in the supply chain, whether it's related to, you know, geopolitical strife or tariffs or COVID-19 or what have you, are the ones that win. The largest companies also tend to be the least nimble. With digital infrastructure, we still are at the stage where the companies are still run without a ton of bureaucracy in them. They're able to shift fairly quickly. What I learned is flexibility is key to being able to react to shifting dynamics, which is a lesson of COVID. [00:26:39] John Wilson: Two things, right? One, one of the importance of having good fundamentals from a procurement perspective. And then the second is, I think it caused people to rethink about the risk management dynamic. And for a long time, sourcing components out of one or two countries because it was the lowest cost. And then that was a way to reduce cost risk into your system. All of a sudden, those aren't available. Having a more diverse supply chain, even if it's slightly more expensive, has long-term benefits and enables that agility and enables that flexibility to go and meet the demand where it is. [00:27:17] Speaker 4: So, if you look back at disruptive events, the challenge has always been on the supply side. If you look back at COVID, if you look back at the financial crisis, if you look back even to the dot-com bust over 20 years ago, the sector has always performed in a relatively counter-cyclical fashion. I've argued in a lot of our research that those three disruptive events were very immediately followed and conducive to the sector continuing to grow counter-cyclical to what was happening in the rest of the sector. The reasons were different, you know, COVID, the shift, massive shift in kind of real life to online life, financial crisis kind of kicked off kind of cloud computing, value proposition and the efficiency of computing, taking that and delivering, you know, a very valuable service and a more efficient way to compute. So, you know, is that going to play out a fourth time in less than a quarter century? Maybe not. The focus, I think, is squarely going to be on can the supply chains, can the building of inventory at the pace and intensity of which it now must be done, can it kind of withstand kind of the headwinds that are coming from global economy, geopolitics? I think we can. The trajectory has always been steady over over two decades of hauling this space. In the meantime, if anyone out there has any spare [00:28:33] David Snow: generators or switchgear that they can provide, it's highly, highly needed, right? [00:28:38] John Wilson: We'll take all of them that they have. Yeah. [00:28:41] David Snow: Well, great. Hey, thanks everybody for being part of the Cool Vector Hot Take, and I hope I can have you all back in the future.