All right.
All right, we're ready to rock and roll.
Yeah, let's do it. Let's jump in a minute early. Everybody awake after lunch? All right. I'm Jon Petersen. I lead the REIT team at Jefferies. I've covered Equinix for, I don't know, a long time. Happy to have Steve Madden and Phil Konieczny with us. I'm going to let you guys introduce yourself. I think you had a REIT compliance disclaimer you had.
Yeah, I'll do the obligatory.
Hand it over to you there.
I'll do the obligatory compliance REIT, which is some of what we'll be talking about today contains forward-looking statements. Please read our SEC filings for more information about factors that could affect these statements. Yes, I'm Phillip Konieczny. I'm a Senior Vice President in the finance organization. I've been with the company for 14 years.
Steve Madden. I'm the Vice President of Global Technical Advisory. I advise customers and partners on their infrastructure, and I've been with the company almost 11 years.
All right, great. Why don't we start maybe just a little more of an introductory question? You guys have about 260 data centers across more than 70 markets worldwide. Maybe help everybody in the room kind of understand how your global presence benefits customers and what strategic advantages that extensive footprint provides.
Yeah, thanks. Yeah, I mean, first of all, it's one of, I think, one of the core tenets of what we see as a real differentiator for us in the marketplace, having the global presence that we have. In addition, we've cultivated these rich ecosystems in most of those metros as well. The third element is really our reliability that we provide to our customers. We see those as all kind of core building blocks of our kind of core differentiation. Yeah, I think the global presence that we've had has served us incredibly well, especially as we have been serving more multinational customers over the course of our 26-year history. If you look at some of the statistics, we've got over 60% of our revenues comes outside of the United States today.
When we look at our customer diversification and how they're diversifying their deployments, over 75% of our revenues come from customers who are deployed in more than one metro. Even when we look across all three regions, we've got 64% of our revenues come from customers who are deployed across all three regions that we are in. It has really been a key differentiator for us, again, as we think about where customers are looking. They're there to architect their IT infrastructure. You throw in things like data sovereignty and other elements which are incredibly relevant, especially in places like EMEA. Having presence in so many markets there has really served us well. Steve, any other thoughts on how companies are looking to architect?
They want to do business in all three regions, but I think we reduce the barrier to foreign markets dramatically. So even small companies can be in 30 countries within a year leveraging one common provider and partner for the ecosystem.
Okay, great. On the last earnings call, you guys mentioned accelerating your build, your development process at, I think, NY3, DC16, and LD4, kind of pushed them up by a year. What are some of the utilization rates and pre-leasing levels at those facilities that kind of caused the push-up? Maybe just talk about the kind of the level of demand that's, I guess, pushing more of a proactive approach to development.
Yeah, sure. I mean, I think this really absolutely ties into what we've stated as our build-bolder strategy, right, which is to build probably fewer but larger builds as we think about things. More importantly, instead of we used to build in more modular chunks in a given metro, now to build that larger footprint in a more capital-efficient way is part of it. It also really is giving us a better ability to serve some of these customers' needs of larger footprint demands that we're seeing across the enterprises. Think of kind of the 0.5 MW-5+ MW type of deployments. They're not hyperscalers. They're not ready for kind of our xScale, but more deployments that where, again, these enterprises are looking to take down more capacity.
Importantly, where they've got growth runway so that they know that they can grow into that. That's really kind of behind this whole notion of building bolder and kind of building in these bigger phases. Therefore, that's what we were kind of referring to when we said kind of pull forward. It's really, instead of building it in two different phases, we're going to do it in one phase in a bigger deployment, really, again, to be able to satisfy the growth needs of our customers.
Okay. All right, great. That's helpful. I guess when you make the decision to build, I mean, how much space do you try to pre-lease ahead of time, or are you building most of it on a speculative basis?
Yeah, obviously, we're different than a lot of the kind of the wholesale community, which does a lot more pre-leasing. We do that definitely in our xScale facilities, right? So when we were 85% of the capacity that we brought online in xScale is pre-leased. The retail world is a little bit different, right, where we're building kind of in metros that we know well. And we kind of have a really good sense of the demand environment and the uptake rates. We don't do as much pre-leasing. I will say, though, now, because of a lot of the power constraints that we're seeing in certain key metros, we are being more proactive about pre-leasing.
Before, we used to look kind of three to six months before a facility would open and start to look at where we can pre-lease that to some key critical customers and magnets for us. We are actually extending that time frame now out to more nine to 12- month time frames where we can get some pre-leasing activity, but it is not like the I thought we are pre-leasing half the building in our retail facilities. That is just a little bit that is kind of not what we need to see and not how we kind of plan.
Okay. I kind of mentioned before you guys are in 70 markets right now, but it seems like you're always entering new markets every year or two. Can you talk about the key factors that drive that decision-making of entering new markets?
Yeah, I'll start. Number one, it starts with our customers, right? Where do our customers need us, want us, and where are they looking to grow? That's the first thing. The other dynamic that's playing out more frequently across the globe now is we're seeing around some of the power constraints. The best example is to look at Singapore, where there are real constraints there. We've been looking at adjacent markets that we can go to, Jakarta and Kuala Lumpur, where we can, Johor, where we can build out additional adjacent metros where there is actually power that we can procure. That's been another driver as well is looking at that. Number one is our customers and where they're looking for and where we see real communities of interest and ecosystems that we can build off of.
Are there any, maybe just broadly, are there any places, corners of the globe where you guys would like to expand further?
Yeah, I mean, we've announced a deal in the Philippines recently, as you've seen, and that was kind of new markets. I mentioned some of the new markets in APAC. I mean, we're always kind of looking at where that next kind of corner of the globe may be, but nothing that's, I would say, a big missing major gap on the map. We just recently have talked about going into South Africa, and I think there could be further activity ultimately as we build out our presence in Africa, but early days on that front.
Yeah. Okay. All right. That's helpful. I want to talk about xScale for a minute. xScale is your hyperscale development business. You guys launched a $15 billion joint venture here in North America late or last year. Maybe can you talk about some of the milestones that you've reached so far? What kind of returns you're expecting and how long we should expect $15 billion to take to be deployed?
Yeah, yeah. Yeah, that's a good question. I mean, listen, we've started off on the xScale business here in America, as you said, with the announcement of this JV. We've procured land outside the Atlanta area. We're calling it the Hampton facility. We're well down the path on that front. I would say we're building a healthy portfolio or pipeline of opportunities. Obviously, you got to start with getting the right land and then powered land, importantly, to be able to fuel the demand that we're seeing there. We're building that portfolio and pipeline of opportunities, and we'll kind of announce those as we start to execute on that. Overall, in the xScale facility, this will bring our total portfolio of xScale business up to over 2 GW of capacity when we look across the entire portfolio globally.
We're excited about kind of our ability to kind of help hyperscalers and continue to grow those communities of interest.
Okay. I guess what are some of the, I guess, risks and challenges on meeting that growth? Obviously, power. We read all the time about power being really constrained. I think the regulatory environment can also be a challenge. Maybe talk about some of those challenges, I guess, domestically and abroad, if you could.
Yeah, I mean, obviously, the power dynamics and, as you mentioned, the regulatory pieces are all the permitting that's got to come along with that and the time frame to get that. What we're seeing oftentimes is not even where we're talking about the power constraints. It's not that we can't see line of sight to the generation of the power, but it's oftentimes the distribution of the power and how it gets to those sites and facilities is some of the challenges that I think the whole industry is for sure wrestling with. That's why you're seeing folks who are going kind of further afield in different areas we may not have thought of as a rich data center environment. For us, we also want to be still thinking about areas that are still kind of near the major metros.
That is critical to, I think, how we think about what our customers are looking for and ultimately how we are building, again, the ecosystems around oftentimes these hyperscale deployments.
Okay. Maybe just as well on that for a minute in terms of the guys that are building far afield that talk to some of these companies sometimes. I guess how do you think about broadly of your strategy of, I don't know, being in North Dakota versus Northern Virginia and even as you build out your xScale business?
Yeah, I think the biggest thing we probably really think about is kind of releasing risk, right, around that. You can sign up one customer and get that deal, get the project obviously built and commissioned. When you think about we're in it for the long term, right? I mean, that's one of the things I think that we've always been saying about the portfolio in our xScale business as well is we're not just developing then to kind of sell it, but really to kind of build that ecosystem kind of around it. That is why we're focused on the metros that we're focused on. Again, thinking about from an underwriting perspective, what that releasing risk really looks like is something that is really critical to us as well.
Okay. All right. Steve, maybe bringing on this. I guess talk about how customer needs are evolving, maybe specifically in xScale, but also curious on the colo business in terms of power density as we're trying to build out these AI workloads.
Sure. I mean, generally, it started, a lot of the infrastructure deployment with Equinix was intercommunication and interconnection and networking, which is very low-dense cabinets, very high connectivity. Of course, now we're moving into the data age and the AI age and the learning models and training, which is very power-hungry and very cabinet-oriented. We have a mix of locations that we have had in the retail business where your connectivity should go here, this other data element should go over here. xScale is once you've gone past a certain threshold, you really pushed yourself into that bucket now, and you need to go there. In all cases, we look at the workload deployment type and the characteristics around what they want to deploy. Depending on the density of what they're looking to deploy is what implementation we would wrap around that.
All of our facilities are prepped and ready to handle us to be able to accommodate those different deployments. If it is around the range of maybe 10 kW up to 25 kW, typically cooling is still just fine. In some cases, you could even do liquid cooling cabinet doors and things like that to offset any facility-required implications. Once you go way above that, it is typically a form of liquid cooling, whether that is going to be done liquid cooling in the cage or liquid cooling on chip. It depends on the implementation. A lot of the newer NVIDIA gear, DGX, and things like that are all self-contained liquid cooling on chip. That is a different infrastructure. It does not require air conditioning and cooling the same way as other facilities do.
We basically make sure that, like I said, we architect for the ability to adapt to what's going to be commonly asked for in that space and then implement the infrastructure required to do that.
Okay. So when you think about building, we talked earlier how the next phase, you do not often do a lot of pre-leasing. I think you are talking about the retail business. How do you kind of future-proof that for where we are headed?
Yeah. If anything, we're lucky in that we have a campus of buildings. You can have one building tooled up, ready to go for the higher density and mix of the others. Again, a lot of our customers have come to the realization that they don't have to have all of their deployment in one cage, in one building where they can see it and hug it. Now it can be sort of distributed amongst different places in the building where it makes more sense, especially when it enables them to grow that deployment. We've had a lot of conversations like, "Okay, it's whatever it is today, but what is this going to look like in three years?" Like, "Well, it's going to be this big. Well, then don't put it there.
Let's go and stick it over here." So it's basically the right workload with the right price point in the right facility.
Okay. I guess I'm curious in the xScale business, we see headlines about some of the hyperscalers delaying construction projects because they're having to reconfigure how they do the cooling and all of that. Are you seeing that as you're doing builds for hyperscalers in the xScale business? Are they figuring it out as you go, I guess?
I think it's not so much figuring it out as we go. It's like where's the tipping point where you're predominantly going to be deploying more of that kind of infrastructure, in which case you might architect more line towards that. A common example of that would be rather than tool how we distribute liquid cooling infrastructure to a cage, every cage has a conduit of liquid cooling infrastructure already there. As they deploy, you tap into it. That's a fundamental shift from adding liquid cooling to, "No, it's going to be liquid cooling by design and backing out." I think you'll find that generally once you reach that certain threshold of either PUE and WUE efficiency, power usage efficiency, water usage efficiency, liquid cooling and that infrastructure is more efficient, it's just more expensive.
As we reach to the scale of deployment where it makes sense to do it that way because you're going to get the return back on scale, that's why we should do it that way.
Do you have a sense with some of the xScale builds, or maybe you can talk about how it's evolved over time, like how much of the use case is just pure AI training versus some hybrid of traditional cloud workloads and maybe some inference and AI? Talk about that, yeah.
Yeah. I mean, I know GenAI and LLM and everything was kind of these bombshell, massive megawatt deployments. There's actually very few of those. Most of the use case is happening in enterprise. Even with other than GPU as a service, which is just pretty providing capacity, I'm mostly around more targeted use case specifics. They're trying to train a model that's going to do building efficiency for energy usage, or they're trying to do visual computing for facial recognition or number plate or license plate recognition. Those tools get trained, but once they're trained, they do not need quite as large a model. You're not training it on the entire internet. You're training it to do very specific things. Once you see the models get developed to do specific things, they start connecting them together and building bigger solutions.
It's not actually going to eventuate that we're going to have 500 or 600 large language models like GPT-4. It's not going to happen. If anything, it's going to be a plethora or portfolio of smaller models that do specific things that are applied to the operational usage that that company cares about. They won't even necessarily have to train it because if it already exists and you just get it from a model as a service company or even from NVIDIA and just drop it in and run with it, you'd rather do that than go and train it all yourself.
Do you have a sense of what the power requirements are of that sort of inferencing workload that you're talking about?
Much smaller. If anything, in some cases, a lot of these models do not require GPUs at all. CPUs are perfectly fine. We have TPUs and Groq and other chip technology service that are very specific to use cases. They are just fine. Also, the infrastructure is distributed now. You actually have infrastructure in different metro cities that is handling maybe a dozen or a couple of hundred stores. You do not have to bring all of that back to a central hub. You are able to push it out and distribute it a little bit. We see infrastructure growing in the edge computing spaces that way.
The computational analysis of it is more, how do I process real-time telemetry or more real-time-based data and make my inferencing decision and less about huge amounts of volumes of data that needs to be used to train the model to understand what it's doing in the first place? It is very much an operational deployment, not an academic sort of deployment.
Right. Which gets back to, I think, some of the advantages we've got being so distributed across 75 markets around the world and where these things start to get aggregated and again, smaller inference-level applications where we can play really well.
Yeah. When it involves humans, like trying to make them more efficient stuff, it can all run from cloud or from the regional centers. Anything that's operational technology-oriented that's not human-related, you see that getting deployed at the edge. Until they deploy it in their operations and start seeing either efficiencies or improvements or throughput or something, they're not getting paid back on what that investment on AI was.
Yeah. Okay. That makes sense. Maybe just staying on this topic. On the last earnings call, you guys do not often name the names of your customers you signed deals with, but you made a point of talking about you signed some AI-related deals with Block and NVIDIA and Groq. Can you maybe talk a little bit more about how those AI deployments, maybe how they differ in design to traditional workloads and kind of what the opportunity is with those kinds of customers?
Sure. Obviously, a lot of it is related to providing pre-deployed, pre-configured stacks for AI development and inferencing as well. In some of those cases, it could be a GPU as a service where they're just simply hosting the capacity in proximity to connecting it to your own data and your own cloud instances, etc. That just becomes part of their environment and they subscribe to what they need. Over time, they may not know what that's going to look like. It does not make sense to buy it all yourself if you're not really sure what the long-term sizing is going to be.
Instead of you can use cloud, sure, but then these other GPUs as a service or another acronym is NVIDIA Cloud Providers, NCPs, can provide you these kinds of services for while you're moving through that development cycle to figure out what your steady state's going to look like. We're finding that because there was a supply constraint and because there's different models that are tailored towards different industries for specific cases, you're going to see a marketplace show up. We're starting to make sure the foundation for connectivity, data distribution, and local government sovereignty, having GPUs as a service available for our customers, and then bring in all the software stack sort of companies, that's when we say ecosystem, it's not just networks and clouds. It's like, who are they actually going to need?
Do we have them in the building ready for them to go?
I mean, it's starting to remind me of the early days of the cloud, right, where initially it was the big hyperscalers building large buildings, then they did all these on-ramps, which they naturally put a lot of them in your data centers. Then hybrid cloud became a huge demand driver for you guys, not just the cloud companies, but the enterprises. Is what you're describing hybrid AI? Is that the right buzzword or is there a different one?
I'm starting off maybe developing it in the cloud, but then I've got to actually roll it out into the field and you need an infrastructure to do that. We are not only helping cloud providers with their space and connectivity, but we brought all the enterprises to them. That's their customer base. We're doing the same thing now with NVIDIA and GPUs as a service as well.
All right. Great. I've got a few more questions, but if anybody does have a question, you can feel free to raise your hand. Let me jump into one more here and then we'll take questions from the audience. You guys have pointed, like in your most recent earnings call, you talked about more substantial sales cycle efficiencies and conversion improvements. Maybe just talk a little bit more about those improvements. Is that structural? Is that temporary? How should we think about?
Yeah. I mean, I think we absolutely think that's structural, right? I mean, we're doing a lot of things internally to try to, we talked about this notion of both serving better, which is how we serve our customers more efficiently and effectively and get them quicker to time for deployment, which benefits them as well. I mean, also just how we run simpler internally, right? I think a lot of what we're trying to put in place now is, again, how do we shorten those cycles, reduce the touchpoints? We talked about, I think Adaire mentioned this on the earnings call, that even simple things like, are we going to be entering into long drawn-out negotiations with deployments that are below a certain size threshold? We've stopped that without exceptions. That's just taken a lot of friction out of the system.
It benefits our customers. It benefits us internally to help get us quicker to revenue generation as well. We are going to continue to look for opportunities, and we see that there are many to be able to continue to streamline our business internally and also to, again, serve our customers in a more efficient way. That is why we think it is structural in nature as we are changing kind of fundamentally how we think about, again, going to our customers and how we are going to service them.
Okay. All right. Do we have any questions from the audience? Yeah. Go ahead.
I was just curious about your cooling that you guys have currently is water-based. Have you considered, and because I've heard of it on this podcast, that there's a new way to cool it, which would be using nuclear energy that was supposedly safer? Have you ever heard of that?
The question is on different cooling technologies and how nuclear energy might factor into that. Does?
Nuclear cooling, I have not.
Yeah. Power and cooling, I think it was both in there. So there's different cooling technologies being used in our facilities today. Clearly, we're trying to move away from ones that consume water. And we're happy to say that our WUE, which is our water usage effectiveness, is some of the best in the industry. The reason why that's important is because power usage effectiveness and water usage effectiveness, you need both to understand how well a data center is doing their job because it's really easy to make power usage effectiveness look good and make this look terrible if you're not reporting on it. Yes, that's a big deal for us. I think what's more interesting is that the industry is driving towards newer technologies like liquid cooling, which doesn't require HVAC and water-based evaporation-based technologies because it's required in order to move that much heat.
The old way we had it does not even work anymore. You have to have a newer way of doing it. We are already in a place where that is what we are saying with the tipping point, when is it typically all going to be more liquid cooling than even evaporative or heat-based? We are seeing a shift as the densities get higher to around, if we start to look at it, if it is over maybe 15 kW or 20 kW, we start going down the path of, well, this is just going to grow. Why do we not architect for that in advance, right? We are absolutely all over that. It is not typically water. That is why Keith is using the word liquid. A lot of them are closed systems too, so it is not consuming liquid.
For the nuclear and power distribution, we're obviously very interested in anything related to energy, alternative energy, and green energy all over the planet. We're exploring different things in different countries. We have some things running already, like fuel cells, Bloom, and things in different facilities, trialing and getting them going. None of them are at the stage where we don't need a utility provider to still provide power. We have looked at what's happening with alternatives like hydrogen, etc. I think the inspiration around nuclear, although its naming, its marketing, is going to be a problem, is that it's actually a very efficient way to create new supply of energy in location to where there's already a problem.
The distribution problem of the grid can be overlooked a little bit, which means you can get the benefit of it relatively soon comparative to building out piping for hydrogen all over the country would take decades, right? If you can reuse the existing infrastructure and solve the problem of the capacity gap relatively soon, that helps not just from our own supply perspectives, but even for our sustainability goals because it is offsetting power we would have used somewhere else. We are very interested in that. We are watching what happens in the market. We are alongside everybody else trying to drive for these technologies to come to fruition, but we do not have control over when that is going to be.
Okay.
Question. Talk about water and power. With turbines, kind of internal measure, fueling with liquefied natural gas, they're very mobile. Can you talk about the viability of that and if this can be permanent or it's just a bridge to connecting to the grid or?
Yeah. Yeah. The question is like the viability of natural gas, some of these more mobile use cases of it.
Yeah. I mean, it's different in every region and even in every country as to what you can and can't do in the first place. To your point, is it even worth it or not? It's hard to say. I think when we look at it, one of the ways I've seen this used is rather than think about it as an alternative to utility, can it offset how much we're drawing from the utility? Or could it help provide a better alternative backup if the utility went dark and we had to run on our own generation of power? How much could that help us there? It's probably like where can we use it effectively? As a mainstream alternative, right now, they're too junior or inferior for the purposes of replacing the need for utility power.
Like I said before, hydrogen is very exciting, but we would have to plumb a giant pipe of hydrogen to our data centers because you can't bring it in on a truck every day. That's just not going to work.
With natural gas, it's obviously plumbed a little bit more in many areas. Listen, we are looking at on-site power generation as both a bridge for sure. There are metros, more in EMEA, where we have deployed that as a technology to get the power there. It's been working great for us. It is definitely a technology we're continuing to look at more aggressively.
That's right. As I said, in Europe, however, not so much. Gas isn't a good alternative. It depends. I think the other thing that's interesting is cooling and finding other ways to disperse heat hasn't maybe got as much press as it should. We've been heating Olympic swimming pools, food and greenhouses and things to try and grow food. We're doing a lot of other things that weren't really on the cards a year or so ago that I think all help and contribute to the end.
I'm thinking of development. I'm thinking of are there any equipment shortages you're suffering at the moment? Are there any tariff implications associated with that?
Okay. Tariff implications and supply chain on materials?
Yeah. I mean, just quickly, I'll touch on the supply chain. Obviously, we've been dealing with supply chain constraints for a few years now. It's something that as being one of the biggest providers and consumers of this equipment, we've got great relationships with lots of vendors. We're well ahead of the supply chain with regards to how long we're ordering it well in advance to make sure when we need it delivered on-site that we've managed that appropriately. Definitely, it's a challenge our procurement team is dealing with on a regular basis to make sure that we've got the kind of the long pole in the tent solved for, right, and figured out. On the tariff side, again, knowing that that's just affecting right now on the U.S. side, and it's very fluid.