Equinix, Inc. (EQIX) Earnings Call Transcript & Summary

Okay. Might as well get started. Thank you all for coming. My name is Cameron McVeigh. I cover communications infrastructure here at Morgan Stanley. And pleased to welcome Jon Lin, Chief Business Officer at Equinix.

Thank you, Cameron.

Welcome. And before we get started, I have to read this. For important disclosures, please see the Morgan Stanley research disclosure website. If you have any questions, please reach out to your Morgan Stanley sales representative.

Okay. Done. Jon, let's get started. Equinix reported fourth quarter and full year earnings in February. CFO, Keith Taylor, called it the best quarter ever. Jon, what in your view was the main driver of this momentum?

Yes. I think a couple of different areas there. I'd say, first and foremost, when you look at the performance of the business in terms of our bookings, which is how we treat our customer sales essentially best quarter ever. And I think in a recurring revenue business, we're always under pressure to do that. But to be able to deliver that in a momentum that we're seeing around that, I'd say, I think for the full year of '25, we end up delivering 27% growth in bookings year-over-year. I think for Q4 end up being closer to a 40-plus percent number compared to the previous Q4 of the prior year. So it really just shows, hey, you're starting to see continued momentum in the business in terms of customer demand manifesting across a number of different dimensions. I'd also say in terms of managing the business, I'd say we're continuing to be -- see continued improvements on driving operating margins. Also, the treasury team has done a fantastic job of managing the balance sheet, continuing to do work on our capital raises, et cetera. So when you put all of that together, I think it just put us in an incredible position to win. And that bookings and that customer growth was also across a really distributed base. I think we had 4,500 deals across 3,400 customers. And so it's not one particular segment that ends up translating into across both service providers and enterprises and across different segments inside the enterprise base, really sustained demand.

Great. And as you just laid out, solid 2026 guidance was well ahead of market expectations in the previous guide given at the Analyst Day in June. Maybe just talk about your key priorities and key focus for the year.

Yes. I think 3 different areas there. First, continuing to capitalize on the customer demand that we're seeing out there, satisfying the requirements. It is definitely an area where, one, we've got to make sure we've got the capacity to be able to satisfy the customer requirements. We've got to make sure we've got the coverage and kind of mechanisms from a sales and marketing standpoint to get in front of customers, hear the requirements earlier to be able to understand and shape the demand as they're thinking about these AI requirements. Over the course of the last 3 years, I think we've been looking at AI at large and our early read around the industry trend was going to be -- there's going to be a lot of activity on the AI training side, which I think we've all seen. There's going to be a lot of investment into that, again, which we've all seen. Our belief was always the durable value creator for AI is going to be around inference. And when that materializes in a real way, that's when we'll want to go ahead and continue to step up in terms of our activity. And I would say that's what we're seeing. I think some of the announcements that we had at Analyst Day around our CapEx intensity and the increase there was a real realization that we're starting to see the enterprise takeoff here around durable use cases, around value creation for them. And it's not just on the service provider side where, of course, you'd expect service providers to invest early to be able to see like build in front of where the revenue capture was going to be. And so now we're seeing that kind of long-term trend across multiple different segments around various use cases, both ones that are going to be agentic in nature as we're all hearing about now around how that's going to change things around generative AI, around also though core machine learning systems across different modalities that are going to need different intersections with different data sources. And we believe our -- again, Equinix' fundamental value is when data is from different locations, whether that's geographic, whether that's in different clouds, whether that's from different counterparties, we're the logical place to interconnect all of that and really bring that to life. And so we're starting to see that demand. So the priority #1 is keep building as fast as we can around that, keep delivering that capacity to support these requirements, keep staying in front of customers to understand not just where are those requirements to land immediately, but also where are those requirements into the future so that we can -- again, you've heard us talk about we're preselling more than we ever have in the past. We want to be able to capture those requirements to inform our development pipeline as well. The second area is just we're continuing to drive improved operating margins for the business. I would say, again, using a lot of the tools that we're looking at for digital transformation that our customers are talking about, we're also seeing the opportunity to transform our lead-to-cash systems, our kind of back-office systems to be able to drive increased efficiency there, and really flatten the curve in terms of the growth of those costs. And then the third area is managing the balance sheet. Again, capital-intensive business, making sure we've got kind of best-in-class cost of capital access to be able to deliver services and build as we're -- again, a lot of our construction can be multiyear in front of when the revenue is going to show up. We want to make sure we have the best balance sheet and structure possible to drive that.

Great. That's helpful. Jon, I wanted to ask on latency. How often does the issue of latency come up when you're speaking to customers? And are there any particular workloads that are latency sensitive? Maybe how do you expect Equinix to serve a greater share of those workloads going forward?

Yes. Certainly, all the time because we bring it up. But you think about it from 2 lenses on for interconnection requirements for our customers, there's both the latency and also the throughput required for these workloads for AI in particular, but across the board. And the way Equinix has decided where to locate ourselves geographically, operating in 77 metros in 37 countries now around the globe, it's to be closer to where the eyeballs are, right? We're not developing our data center campuses where there's cheap power. We're not like looking for empty farmland. We're actually located where the nexus of population and like fiber connectivity exists so that we can serve those workloads. And so whenever we're out there, we like to say, I think we're within 90% of the world's population within 10 milliseconds, right? And so when you think about 10 milliseconds, it's faster than the blink of an eye, it really lets you serve almost any kind of workload that is latency sensitive, whether that's real-time interactions, whether that's gaming, whether that's financial trading, whether that's media distribution most effectively, there's a lot of different areas that we can drive around that. And so those conversations come up a lot. I'd say for AI, in particular, it ends up being twofold. One, when you're -- when all of us are interacting with our chatbot of choice, like that's not particularly latency sensitive, right? We're all waiting for the dots there and like the token generation takes longer than the latency of the network. Where it does become valuable, though, is when you start actually dealing with multimodal systems, so different data pools that are coming in there being ingested by your AI, well, every time that machine needs to go out, establish a new connection, find a new data source, get that and process that, well, that's like essentially hair pinning that traffic back and forth. And so when you start seeing those effects, that's when you really start to see latency sensitivity come into play. The second is any kind of machine-to-machine communication. Again, when you start interacting with humans, we're all slow, that's okay, especially on a Monday morning. But when you start seeing the machines actually trying to converse with each other, if you're not able to have the lowest latency possible, well, that's just costing you either like time to token or time to value, whatever your transaction might be. And so in the case of financial services, that might look like, hey, for risk management calculations for end of month runs, that's not particularly latency sensitive. When you're trying to do trading and you're trying to make sure that like you understand what your next decision is going to be on a purchase around that or you're trying to go ahead and inform what your next set of currencies that you might want to move during a day period might look like around that or commodities today, right? Like any of that, the latency sensitivity can be dramatically higher. And so like that's an example of a workload. But I also mentioned like in addition to latency, the throughput has actually become a more and more important element of the conversation. And like for a long time, we've been dealing with data across the network is increasing. But I would say post-COVID, it had kind of plateaued a little bit, right? I mean HD content is HD content is HD content. 3D-s didn't come about. 4K like streaming is there, but it's not massively relevant in terms of passing this. So the part that we had, as an industry, been thinking about was like what's going to be the next big driver of traffic across the network. And I would say AI is an example of that, right? The ability to process AI is entirely reliant on the ability to process data and actually like process those workflows quickly. Every time that, that data is hung up on the network instead of hitting your GPU, you're costing yourself money, right? The GPU is the most expensive thing on the planet right now in terms of bringing that to life. And you'll hear Jensen talk about that tomorrow, like the unit of value that is most important is making sure that these GPUs are maximized. And that's why we're focused so much around like, well, in order to maximize that GPU, you've got to make sure that data is flowing into that thing as quickly as possible. So that's increased the amount of throughput required. So we have -- for Equinix, a software-defined network that we call Fabric. It connects to both the cloud and all of our locations together. When we started that, those were 1 gig circuits, growing to 10-gig circuits with like 10 gig ports, growing to 100 gig ports. And now it's like we're upgrading that network fully across the network to 400 gig ports with 100-gig DCs, right? And so a 10x multiple in terms of the throughput and capacity required. And that can be anywhere from our service providers that are looking to drive AI traffic and workloads into enterprise deployments in private clouds to, again, enterprises trying to move data and workloads from different clouds into a central data repository so that they can have their private AI work against that.

Great. Related to that point, during your presentation at the Analyst Day, you mentioned partnerships with NVIDIA, Dell, Groq. How is Equinix leveraging these relationships? How has the partnership evolved? How do you expect these partnerships to help capture more of these AI workloads?

Yes. We're in a unique position in the marketplace as Equinix, where like we've talked about the neutrality of our business, and that's been core to our thesis since we were founded 27-some years ago now. And neutrality means a couple of things. One is we're carrier neutral. So we invite all of the different carriers to come to our facilities and exchange traffic, but it also means we're technology neutral, right? So we want to make sure that we understand and can bring to life all of the most important technologies in the world to our customers. And that's both from a sales perspective, but more importantly, from how can we accommodate that and how can we bring that to life for these deployments. Our partnership with NVIDIA started maybe 5 years ago, right, before AI was a real thing and before enterprise AI became a real thing. But what we saw was, hey, they're making incredible progress around machine learning, and that's going to continue to be more and more relevant. Well, when that infrastructure comes to life, how are we going to be able to support that for our customers in our data centers. And so a lot of our conversations start there. So they've ended up evolving and obviously becoming an incredible partner from a go-to-market perspective. But first, we want to understand how can we help them bring this technology to the rest of the world's customers. Again, our customers are the world's digital leaders at 10,500 customers now, roughly divided, 50% of them are service providers. So we've got all of the major cloud providers, all the major networks and all of the emerging ones, the neo clouds that are developing and all of the enterprises that then need to consume those services. So it's really important for us to then not just understand, but then also be able to take an opinion to these customers like when they're asking how can I support these liquid cooled GB200s? I have no idea. I don't have liquid cooling in my enterprise data center. We can tell them, oh, we've already thought this through. We've worked with NVIDIA for 18 months on designing solutions around this. We can operate this for you with the same reliability of any production facility tomorrow for you, right? And you can go ahead and start deploying. And so that comfort level that can come from that, but also the value then that they can get from that technology becomes very powerful.

That's helpful. And on the topic of liquid cooling, curious if you could provide an update on the rollout of these liquid cooling solutions in the new and existing capacity. And then just how important is liquid cooling for AI inference?

Yes. So Equinix has about 280 data centers around the world. Around 100 -- slightly more than 100 of them at this point across 45 metros, again, globally, are liquid cooling like able to support, right? We've done pre-engineering work, and these include facilities of ours that are 10, 15 years old, where we understand what the central plant looks like. We understand how to bring water in cooling systems directly to a customer infrastructure, and we're actually comfortable operating that, right? I'd say 3 or 4 years ago, if you asked a data center operator, oh, we're going to have water coming to customer equipment that's like heavily electrically loaded. Is that going to be okay? Everyone is like, that might be pretty challenging for us. And so we spent a lot of time figuring out, well, how do you actually operationalize that, right? The plumbing is one piece, but knowing what to do in the event of a leak, knowing what to do in the event of a fire, knowing what to do in the event of an operational situation where if you lose cooling capacity into that liquid, these machines are so hot that you will thermally overrun essentially within seconds, like that's obviously a more complicated scenario, right? So to be able to do this in production and be globally consistent around that. So we did that work about 2 years ago, right, in terms of understanding how to productize the capability for liquid cooling, bring that to life for customers and have been solving for that in what we would consider like a standardized, productized way. We had been doing this as one-offs though for customers for over a decade, right? And so like customers would ask us and we try and figure that out. So that is going great. I'd say we continue to see customer demand scale now, which is super exciting to see. I think we built honestly, in advance of where those requirements were going to be. And now we actually understand that pretty well. We're doing this in a repeatable motion. It's still, though, from a like total percentage of deployments or like customer request basis, it's still the minority, right, to be clear. It's -- when we see these requirements, again, in the 3,600 transactions that we talked about there in the like low double-digit percentages of total deployment. So call it, maybe several dozen to maybe 100, we'll have inquiries around that and figure out how to support that. So it's still relatively low volume. But these workloads are generally -- when they are liquid cooled, it's almost always going to be an AI deployment. That's probably the most expensive gear that's on the planet that's going into those deployments. And so we want to make sure that we're satisfying the customer on that delivery.

That's great. Jon, I wanted to hit on cabinet density. Maybe just an update on your power density across your existing portfolio versus some of the planned data centers coming online, and how we should think about the average power density currently and at what rate that might start to tick up over time?

Yes. It's an interesting one. Obviously, you look at like Grace Blackwell, and that's at 120 kilowatts of cabinet for liquid cooled deployment supporting the latest NVIDIA, and that's going to go up, right? We all know like NVIDIA is talking about 300 kilowatts of cabinet for future architectures around that. And that's true. Like, again, the deployments that we have supporting some of that are at 120 kilowatts of cabinet. But when you think about the cabinets that are also supporting that, right? So it might be 4 or 8 cabinets of 120 kilowatts a cabinet GPU capacity, it will need 20 cabinets of storage. It will need another like 5 cabinets of networking gear and infrastructure to support that. And so that average can actually come down pretty rapidly. Across our new deals, I think new deals booked were at 6.6 kilowatts a cabinet across the entire Equinix fleet. Like our existing base, though, across all of the things that are deployed and live is somewhere just north of 4 kilowatts a cabinet. When we think about data center design for the future, we're certainly upticking that though, right? And like the latest design that we're deploying in Dallas, I want to say, is around 18 kilowatts a cabinet. We generally think somewhere between the 15 to 20 depending on the mechanical infrastructure that we have in place. That feels like the right average. And the reason I say that is we'll have liquid cooling to support as dense as you want. I think the latest designs where the team was telling me it's more like 500 kilowatts of cabinet or maybe even a megawatt. It's just you need a bigger pipe fitting for that, which blows my mind. But the important part is it gives you the flexibility, though, right? Again, if we used up, let's say, it ends -- let's say we're wrong on density that ends up being overly dense. We can support that with the liquid cooling infrastructure that we've deployed there. All it means is we strand shell, right? And honestly, the shell of a data center is the cheapest part about the construction. We want to maximize the utility feed that we're getting out of the utility provider. We want to maximize our electrical and mechanical systems. So we're delivering that. If we're wrong the other way, if you build too dense and you actually can't sell that dense, you've stranded all of your capital investment on all of your electrical and mechanical plants. So that's a really bad outcome. And what we've seen over the years is it's really hard to tell where the puck is going to land. Our data center investments when we're building them, this is not a 3-year investment, right? This is a 2-year construction project, and we're expecting that to be able to deliver revenue for us for the next 20-plus years. Like we want to make sure we're thinking about flexibility of that infrastructure then and the long-term value that, that thing can drive for us because we're not just worried about the customer for today, worried about the customer on the renewal. We're worried about the computers and workloads that will be supporting there. And what we've seen over time is that can move, right? Again, everybody has been focused on GPU density and like the training in that density. It is certainly very dense. When we look in -- and we've been talking with silica start-ups for the last 5 years, when you think about inference and even like NVIDIA's acquisition of Groq, the AI inference chips are actually much less dense, right? The compute required to drive that, the complexity of the silica is significantly lower. And when you think about inference, a lot of that can be air cooled, that can be lower density. And 5 years ago, everyone was saying, hey, Arm is going to be the -- replace all of x86 chips in every data center, densities are going to go down to 4 kilowatts a cabinet. And maybe that will happen, right? So we want to make sure we're designing though so that we can go ahead and drive revenue and value for the long term.

Got it. So I'm curious how you think about price elasticity of your customer base. Monthly recurring revenue per cabinet has been going up. And would you expect this to continue over 2026?

I think there's definitely some amount of price elasticity there. I think it's certainly something that we're always trying to identify and figure out, hey, where are the opportunities for us. And it comes from a couple of directions. One is our pricing per cabinet will go up like inflation is a real thing around the world, right? And like our input cost in terms of data center construction, development, all of that, it has been going up over time and probably will continue to. Utility costs are highly variable. That can go up over time. The density of the cabinet itself, we're delivering, again, more kilowatts into a cabinet than in the past. So that can drive pricing and yield up on a per cabinet basis. Our interconnection offerings continue to drive more and more value for our customers. And again, one of the reasons Q4 was so great for us, like when Keith says it's firing on all cylinders, like interconnection actually started to reinflect in terms of growth, and we're starting to actually see that go up again, that drives more dollars into that yield. So I'd say, overall, it's a complex algorithm to be able to get to exactly how far you can push that. But I will say the part that's very like interesting over the course of the last like 2 years, especially around AI and the importance of the workloads that are being driven there, the value creation being done out inside of that infrastructure now and also the pricing of like the GPUs and that technology stack is significantly higher than your standard like web servers of the past, right? And so when more and more value gets created out of that deployment, then there's a lot more elasticity around like, well, it's going to cost more to make sure it relies -- runs reliably is always on and can be connected to all of the data sources that matter like people are like, okay, I'll pay up to be able to go ahead and make sure that I'm maximizing my dollar value out of this massive capital investment that we're putting in.

Great. Jon, I wanted to be sure we mentioned power and talk about it. On the last earnings call, Adaire had mentioned the 3 gigawatts of developable capacity is powered or close to being powered. Can you talk about maybe just the power procurement environment and potential challenges Equinix is seeing at the moment?

Yes. It's -- on a global basis, it's interesting and like for the first time to be able to see in the mass media like talk about data centers all the time. So infrastructure is sexy again, that's great, except when it's negative press. It's fascinating, right? I'd say, one, you have to be long-term thinkers in this space to be able to have the relationship with the utilities that matter, and they have to trust you to know that you're going to be there. And candidly, there's a lot of money that's come into the data center space. They aren't long-term operators. They're putting a lot of weird requests in. And as regulated utilities are, they have to respond to those, right? They're not allowed to just say like, I have no idea what you're talking about, and I've never seen you before, I'm not going to give you power. They have to respond to that. And so that creates a lot of challenges for the entire market of like, well, how do you sort between all of these different requests that may actually be at the end of it with one customer underneath that load, right? And so you end up seeing gigawatts worth of power requests coming into one particular market or jurisdiction. And again, many of us in the industry are like, oh, we know exactly what hyperscalers are, that customer wants that requirement. It's not actually like 2 gigawatts, it's maybe 500 megawatts, and there's this many people that are all trying to sell to that one customer, trying to get power certainty on land. So we've been working with the utilities, and I'm really, really happy with the work that we've done there and just say like, hey, like raise the bar, right? Like actually force people to make it a little bit harder to get access to the utilities actually put processes in place that require like payments for these load studies. So it can't just be random land developers doing that. We want to make sure that there's real long-term providers that are credible making these requests so that we can get some of the noise out of the system. Again, do it fairly. That's obviously very important for us, and we're very serious about our neutrality. Like we want to make the rules for everybody the same, but like just raise the bar a little bit so that you can get some of the noise out of the system. And that's been a great help. And then the fact that we're willing to do that, and we're willing to invest in all of the development, and we've been very like loud and open with the communities at large, along with the utility providers like the data center industry needs to make sure that we're paying the fair share of what we're doing, whether that's net new generation creation or transmission costs that we're creating so that we're protecting consumer rate payers for paying for what we're doing as an industry, right? Because -- and one of those is just from an ethos basis, we take that very seriously. Again, we're going to be long-term owners and operators of these facilities. Our people live there. I live in Northern Virginia. like I see the data center development. I want to make sure like my kids are having the benefit of the tax revenue, et cetera, around that. And so we're going to be in there for a long time, and we're really deeply partnered with the communities. But the second part is like if we look like more selfishly around that or more commercially, if we're doing bad things and consumer ratepayers are like ending up being impacted, you're going to get regulated. And like that's not a good outcome and you're starting to see the pushback now, right? Because again, I would say some players that were working in this industry weren't thinking as long term around that and you start seeing people say like, we should put a moratorium, like these data center things are going to like impact customer housing. So I would just say it's an opportunity for the industry as a whole to kind of grow up, take this seriously and work with government, right, work in close partnership to say like we need to do this. Like one of the good things that I'm seeing out of the administration right now is like seeing, hey, there seems to be that lean of -- we need to make sure we're protecting the consumers from the impact of this net new piece. We've been working with whether it's ComEd or PG&E, you might see some announcements that went out over the past couple of months where, again, because we're working with the utility around this and consuming load of generation that was already there, it actually reduces the consumer payer's price, right? And again, we're paying for substation development. We're paying for actually upgrades to the transmission system to be able to take that. If you have unused generation that's out there in the system, well, the consumer is paying for that as well, right, in terms of a higher unit cost on their power. And so the ability to have that load is useful. You're starting to see more and more happen in the space, though, right? And I think there's a lot of talk right now about, hey, if you're going to bring large load on to portions of the grid, you've got to go ahead and make sure that you're actually kind of bringing your own electrons, so to speak, creating net new generation around that. And like total believers in that, right? Huge believers that we need to make sure that this industry as a whole is paying our share of the way around that. And whether that's ourselves or our customers around that, like the total economic benefit that we're driving around this is high enough that we should be able to take that. And I think that, that's a really important message to communities at large. When they hear about data center development, and I'm sure all of you guys have been covering something like that and watching the space, there's a lot of push like what are these things actually doing? And it's like, yes, the data centers are not producing like cat pictures and stuff like some are. But really, like it is the fundamental like driver of economic activity now, right? Like digital infrastructure is the factory of today. It's the economic -- it's the mall of today. It's the economic center that's happening because so much is happening inside of these facilities. And so it's really important for us as an industry and for our investors, et cetera, to understand that and be able to tell that to your stakeholders around like, hey, like this is actually how like GDP is created now, right? It's a meaningful portion of that, that's going to show up digitally. That's what's happening in the data center, right? It's actually coming online, converting power into money for somebody. That money for power is actually not for the data center developer generally. It's for the customers inside that data center.

Great. And just a follow-up on that point. How does Equinix internally, how are you thinking about using grid connections versus some of these behind-the-meter power solutions, you think about fuel cells or turbines? How does that fit into the strategy?

Yes. I mean we're operating, like I said, in 77 markets across 37 countries. So it's really all of the above strategy. In some jurisdictions and again, in the vast majority of the time, we want to work with the local utility, right? Like grid power is the best solution because you end up creating more shared resource across a broader set of constituents to be able to do that. In cases where the grid is not able to accommodate our power requirements, so we have done and actually created like net new power generation. And we're looking into the future around that as well as like, hey, how can we scale that up? How can we do that? It's -- on a global basis, post kind of World War II and kind of the industrialization of the global economy, like we've actually generally seen power utilization and draw go down, right, for -- I want to say like a 40-year period. And only recently is it starting to inflect back up because of electrification of the grid, electrification of vehicles, more efficiency around heat pumps, like all of that's driving electric utilization up and then data centers as well. And that was like before AI, like you were actually, for the first time, actually seeing like draw increase across the globe. And then now you layer on, hey, all of a sudden, there's also all of this data center activity associated with AI and the electricity consumed by that. So that's net new like load that's coming on. And for a long time, again, because utility was going down, well, those power plants were still there, right? You hear about all these like shuttered power plants. It was like, well, they were created because we were smelting aluminum. And I will say, from an environmental impact to data centers far, far better than like having aluminum smelting in the community you're operating in. And so like that's kind of the replacement that we're talking about now, just like, okay, it's fundamentally converting power into financial benefits.

That's helpful. We have a few minutes left. I wanted to open it up to any audience Q&A, if there are any questions. Now you can think about them. I'll ask one.

Go ahead in front -- there's a mic running here.

Just want to -- if you could just talk a bit more about the 20-year life of the assets and how you see that playing out? What do you -- obviously, the CapEx you've spent, a 2-year build cycle and then you're talking about the 20 years. In terms of the upgrade cycle you expect to have to do to make that 20-year life cycle work for the assets. Can you just talk a bit more about that?

Sure. Yes. And I would say it's 20-year plus, right? Again, most of our assets we actually think about as essentially perpetual. Like we've constructed the data center. We're putting in the maintenance CapEx to keep it going there, but it's going to be generating revenue for the entire -- like outside of our planning horizon, I guess, is the way I would say it. Some of that does require maintenance CapEx, right? Obviously, we're putting that in on an annual basis. There's also requirements like for some of the longer lead elements of that, like we started categorizing like redevelopment CapEx to say like, hey, if we have to replace the 15-year plant, like some of the core data center infrastructure, whether that's the generators that are running or whether that's like some of the electrical switching gear or some of the central chiller plant, that could be like a 15-year asset, and we might be able to say, okay, we can put enough maintenance in there to get a little bit longer out of that. But at some point, you're going to go ahead and want to refresh that. The plus side of all of that work is there's been so much additional work that's happened in terms of efficiency gains and like just upgrading and understanding of how to do that mechanical and electrical work more efficiently that we can upsize that, right? And so in a lot of cases, we've ended up taking like -- take our DC2 facility right now that's going through some of this redevelopment, we can add incremental power availability into that site from both efficiency and from upsizing the infrastructure, which means we can actually generate more revenue into the same facility, right? And so that's how we think about that from an underwriting perspective. But yes, I mean, we're essentially continuing to model our ability to generate revenue out of these things for a very, very long time.

Could you briefly touch on xScale? If you think about it 5 years down the line, has it become a bigger part of your overall business compared to how you thought about it a year or 2 ago?

Yes, it's a great question. I'd say yes is the answer. And certainly, that's the reason that we -- when we started xScale, we were thinking, okay, it's -- our customers have been asking for it for a long time. For folks who don't know what about xScale, basically, our core business for Equinix on the data center side, they're multi-tenant facilities that we call retail, right? In any given facility, we'll have anywhere from dozens to multi-hundreds of customers inside one data center. A large part of the market, like the hyperscalers and some of the other large customers may want an entire facility for themselves, right? And so when we look at that, our average expectation on returns across where we're developing capital for our core business, it's in the -- I think our target that we've said publicly is 25%. It's in the 20% like mid-20s and up range, right? And it's like, again, it's a beautiful business. I'd say when you're developing a single-tenant facility for a customer, like your costs are pretty well known, right? It's pretty apparent to the market at large around that. And yields around those facilities are generally, call it, in the high single digits to low teens. You can take some leverage on that, work that up into the mid-teens. So for 2 reasons then, Equinix had not wanted to pursue that as one of our growth vectors: one, it can be highly capital intensive with a much lower return; and then two, from an actual like the structuring of that because you want to apply so much leverage to that, we thought we wouldn't be able to maintain or at the time when we were thinking about it, get to investment grade or maintain investment grade, if we were doing a bunch of hyperscale development around that. And that's largely been true, right? When you look at outside of the data center landscape, I think there's 2 maybe like depending on how you squint, maybe 4 public players right now. And that's the reason why, right? And so when we looked at it, though, our customers were asking us loudly enough for this product that we said, okay, let's see if we can figure out how to do this. We ended up creating a development joint vehicle structure. Our first one was with GIC, the sovereign wealth fund in Singapore, then we did one with PGIM. Now we announced one recently in the U.S. with both GIC and CPPIB, the Canadian Pension Investment Board together for $15 billion in the U.S. So that's where we're doing kind of solving for the hyperscalers around that. I'd say -- so yes, in the sense that our initial investment for xScale was we had gotten $7.5 billion of equity commitment there. We've tripled that in total size now with the equity commitment from CPP and GIC, and we're executing against that. And I'd say it's delivered what we were hoping out of that, right? One, it gives us capital efficiency, but still gives us the customer -- being able to solve for the customer demand in the ways that we wanted to. And it's also given us quite a bit more scale in terms of our procurement capability on utility as well as equipment. So it's been good, and we'll continue rolling forward with it.

Excellent. We are out of time. Jon Lin, thank you so much.

All right. Thanks all.