Thanks. Okay, good afternoon, everybody. Welcome to the Goldman Sachs Communications, Media & Technology Conference. My name is Jim Schneider. I'm the telecom and data centers analyst here at Goldman Sachs. It's my pleasure to welcome Digital Realty and CEO, Andy Power, with us today.
Thanks for having me.
Welcome. Maybe starting with a kind of persistent topic, maybe even a can't-get-away-from-it topic at this conference, which is AI. So maybe just, you know, it seems to me like we've kind of got this situation where the data center market's got very strong demand, partly fueled by AI, and very constrained sort of power and other dynamics to it. So I wanna unpack those elements a little bit, maybe starting with the supply side. You know, in your Q1 earnings call, you mentioned working with Dominion to help address bottlenecks in Ashburn, and that you were kind of cautiously optimistic about getting access to more power late in 2025, and this past quarter, you talked about some of the constraints easing in Northern Virginia and other markets, too, by 2026.
So maybe summarize for us the current state of play on the power supply side as you see it now.
Sure, so we're operating across 50 metropolitans on 6 continents, so we've been seeing this evolve for some time now, and I'll go into Northern Virginia, Loudoun County, and Ashburn in particular, but I would say this phenomenon is not episodic. Pre-AI demand trends really unfolding in data center, we were basically just running hot for a long time, and we've just started, whether it was digital transformation, cloud computing, IP outsourcing from on-prem to off-prem, basically running into, call it, more and more roadblocks on supply, and Ashburn, Virginia, was the pinnacle of this, or canary in the coal mine would probably be an understatement, given how important it is as the largest market in the world.
But over two years ago now, basically said, "Oops, we're out of power for several years." On the backs of that, having been building, owning, operating data centers in the market for years and years and years, we worked as a partner with the utility companies, looking, call it, at our infrastructure, our substations, our idle capacity, where we could reroute infrastructure and power, and make sure that all our customer commitments were met, and to free up some incremental megawatts that we could sell to new customers that needed that growth in a dire capacity. We also, most recently, essentially granted an easement for Dominion to lay in the Mars Substation, which is a landmark, a crucial piece to the transmission constraint. So we've, I think, been a good partner to Northern Virginia in coming to the table with solutions.
Time has also passed. We're getting closer and closer to the end of twenty twenty-five, the beginning of twenty twenty-six, when the "bottleneck is easing." But I wouldn't say these problems are going away forever, because the demand has continued to remain robust and other bottlenecks have popped up. Substations are a critical piece, and the components or the switchgear, and just weeks ago, there's been expressions that the timelines for new incremental deliveries in that market could be pushed out incremental, call it, twelve-plus months versus prior standards. So I think that what you're seeing here is that the broader, call it, supply for data centers is just gonna be a more prolonged and need to be more thought out in the coming years.
And this is, again, not Ashburn, because this phenomenon is appearing in Santa Clara, just down the road, with even further out gaps in power. There's moratoriums in certain markets, like Singapore or Amsterdam experienced this recently. Dublin has had power generation transmission issues. And it's not just about the power either. It's about sustainability concerns, broader supply chains, nimbyism, a whole host of factors where I think the bar for this industry has been increased. And then AI has kind of arrived at a coincidentally similar time. I wouldn't say AI was the root cause that broke the straw on the back, but it certainly has made the, call it, supply-demand dynamics shift in the favor of providers.
Mm-hmm. Maybe just broadly characterize, like, are there still markets that are getting a lot tighter, and are there any of them that are getting looser?
There's definitely markets that are still getting tighter. Markets that you wouldn't say didn't feel like they had constraints much at all, like a Chicago or Dallas here in the United States, are certainly edging in that direction. Outside the United States, the London market is feeling power constraints, some of our other major FLAP markets in Europe. Asia Pacific's been at the pinnacle of this because it's tougher to build in some of these markets, like a Tokyo, for example. Loosening up, I mean, one case you could say Ashburn is loosening up because we're getting close to the first, call it, big connection to come.
Yeah.
But at the same time, I don't think that it's gonna be a cure-all, and we'll be back to normal in how we do business in that market. So I wouldn't say any are dramatically changing to the loosening side.
Yeah. Fair. Then maybe from the demand perspective for a second, you know, for those investors who are sort of a little bit newer to the space, can you maybe kind of explain? I mean, we've seen strong demand for a while, and all the places you mentioned, can you explain why those markets are the ones where demand has been strongest? And how concentrated is that demand among specific kind of customers and specifically hyperscalers?
We focus on major metropolitan areas, I mentioned 50 around the world, where we see robust and diverse customer demand for hybrid IT, multi-hybrid cloud, hyperscale cloud compute, and now AI is in the mix as well. These markets have had two things going for them. One, they were the origins of the internet in some regards, and that snowball effect has just brought more infrastructure, more fiber connectivity, more access to power, and more customers landing there. And then compounded with the advent of cloud computing, where locational latency sensitivity arose, and the clouds went and picked major markets. They picked their availability zones with radius restrictions, and that has essentially made these markets as desirable as they are today.
AI, which I would still describe early innings of a long process here, the first innings training are not necessarily latency or locationally sensitive. But when we hear our customers' wish list of what's important to them for AI today, being hyperscale customers, large contiguous capacity blocks. Right now, like they're running to the grocery store to pick it up. And they also use the word fungible markets, which goes back to these core markets, because they're figuring this out at the same time, and they don't know if they get this wrong, they don't want to be essentially building this large infrastructure investments in markets where they can't backfill with cloud compute. So that's why I think it's an additive item to the strength of the core markets.
Mm-hmm. And if you go down to the next kind of second tier, third-tier markets, are market conditions there, like, appreciably better? Or are there places where you might see a bit of a spreading of demand into those markets, or are any of those ones actually loosening up?
You're seeing a spillover effect to markets that maybe we had a position in data center, but it was much more network-oriented deployments, colocation, enterprise-oriented deployments, and now hyperscale is becoming a bigger piece of the puzzle. But these are markets that didn't have the benefit of being supporting data center growth for 15 or 20 years, right? So they weren't necessarily ready for the monsoon, and all the ingredients to bring on this digital infrastructure: power generation, transmission, labor forces, construction, operational talent, isn't necessarily ready to absorb it. So some of these markets have popped up, and they're like, "Oh, we're out of power already," real quickly. So there's definitely a spillover effect. A market like in Atlanta has benefited from the current facts and circumstances. There's others as well.
But I've talked to probably more power companies in the last eighteen, twenty-four months than I talked to in my prior eight years at Digital Realty. And it doesn't seem like there's easy power data center infrastructure, the whole kit, where customers really want to go, at anyone's fingertips.
Yeah. I mean, going back to the topic of AI for a second, I think investors, most investors kind of perceive this as something that's happening mainly in a hyperscaler's own data center, data center facilities, so far at least. In your data centers, you know, what have you seen in terms of direct AI demand for workloads, and is that more weighted toward training inference? And when does it become a more material part of your demand profile?
Yeah, I don't. We're not seeing the customers push the AI workload towards a self-build versus a leased capacity. It goes back to those ingredients that I mentioned previously, and most of the customers were already behind on their self-build plans, not ahead of the game.
Yeah.
So it's not like they were sitting there with these idle, large gigawatt capacity blocks and big, vacant shells that they say, "Yeah, that's a perfect place to put AI workloads." I think this is a resource and bandwidth, and you have the key ingredients that I don't have, the customer environment, which is actually pushing them more towards the outsourcing. We had a record first quarter, 55% contribution from what we deem as AI workloads. The hyperscalers certainly led the way in that contribution, probably closer to 25% in the most recent quarter. Again, big contribution from hyperscalers. We are starting to see the tail end of, call it, enterprise-oriented workloads, smaller. They're megawatt, half a megawatt.
Right.
And listen, this is, this has been our bailiwick. We've been pushing the power densities, given our heritage of coming from the hyperscale, moving towards the enterprise colo. So we've been there with high-performance compute, and we've been a great home to win those new applications today.
Yeah. You've been leaning into innovation with your partnerships with leaders in the industry. You know, that's continued with your partnership with NVIDIA on the DGX-Ready solutions for colocation. Can you highlight any specific use cases where, you know, customers or clients have successfully deployed that kind of solution? And how are enterprises actually using that kind of solution?
So with a, with a flagship like NVIDIA, who's now a household name, we were DGX-certified years ago, probably well before the customers were even ready or using it. I believe we're first out of the gates in Japan and Tokyo, I think, ready for the H100s. That was almost two years, coming up on almost two years ago. We had a great win. We are building with NVIDIA, but which will be the largest supercomputer in the Nordics for the Novo Nordisk Foundation, which is a great milestone win of that infrastructure. I still think the adoption by enterprise, especially of the maturity level of that name, we're still in nascent territory here.
Yeah.
Right? So lots of people are thinking about it, thinking about planning their data center infrastructure to be ready for it. But it's the minority of the activity we're seeing today. And fortunately, cloud computing is still growing tremendously with us. Fortunately, enterprise using multi-hybrid cloud, private cloud with us, still growing tremendously with us. That's been the lion's share of the record new logos adding to our 5,000 customer base. That's been the lion's share of the less than a megawatt bookings that have been consistently growing quarter over quarter, year over year. So I think it's gonna take some time for this technology to build out, but I think we have a big hand to play in supporting it.
Yeah. Maybe one last kind of strategic level question for you for a second. Many of your public peers, former public peers, have gone private over the recent years, and in some cases, you know, because of their ownership, and they have access to a significant, quote, unquote, "private equity backstop" in terms of amount of capital available to them, for new facility expansion and the like. What advantages do you have over them, in terms of remaining public, and what disadvantages might come from being public?
We really think about the hyperscale business and private capital kind of hand in hand.
Mm-hmm.
And I think it's apropos as to what comes next with a lot of those platforms, and the rumor mill comes out every day almost now. So obviously, in the public format, we've been very direct and specific that we are focused on compounding, accelerating per share growth, right? In the hyperscale arena, where you're spending significant amounts of dollar in longer and longer builds, so the project's getting bigger and bigger, you have to sacrifice that near-term return for value creation out there in the future. We obviously are doing this in a leverage point that's investment grade, so one-third debt. They're kind of doing it the opposite playbook.
Yeah.
Now, interest rates have changed dramatically in just the last couple of years, so it's the money's no longer cheap or free for some of these cap structures. The way Digital's really tackled is we want to basically have our, call it, fishing pole and all the pools of capital, so that when it comes to hyperscale, we can dial up private capital in various partnerships to help fund that growth for us and maintain the piece of our business that doesn't have that long development drag, that doesn't have the weaker pricing power on cash mark to markets or rental bumps or escalations, i.e., our enterprise colo business as wholly owned, and then utilize our private capital portion business to help, call it, accelerate that bottom line growth.
I'm not sure one's better or worse. They're different, and we're trying to harness the benefits of public capital that lets us raise. We raised nearly $900 million of euro bonds in less than 24 hours on an unsecured basis at a sub-4% coupon just a day ago, or raise equity efficiently as well, but also turn to private capital when we need to use that at the same time.
So tools in the toolbox?
Yeah.
Yeah. Okay, maybe some technology trends that may, you know, Chris is here, elsewhere today, but I want to ask you maybe, you know, power density. You know, beyond power from the supply perspective, you know, a lot of these GPU-based configurations require much, much higher densities. Sort of, you know, previously, we're talking about five and 10 kilowatts per rack. Now, with the H100 and the B100, we're talking about 60, even 80 plus kilowatts per rack. So, you know, your operations, broadly speaking, are you seeing cannibalization of CPU computing by GPUs on a wholesale level, or just in sort of pockets?
I mean, just go back to last quarter, which wasn't a record, but it was a pretty darn strong quarter, and 75% was, we would say, was not AI related, so obviously much more CPU orientation. We obviously have the bellwether in the GPU category right now, pushing the envelope to the most extreme potential in builds. There's still a tremendous addressable market that says, "I'm gonna my infrastructure doesn't need that for that application," right? So, we've made sure that our, our infrastructure has the modularity, the flexibility, fungibility to ramp up power densities as needed. That's often done in retrofits, which we've done historically, which we've done in the last year, which we're doing right now. We've also, call it, pre-positioned our new designs to be, call it, liquid ready.
Quite honestly, some of these customers, they don't even want to use it for GPUs. Their first iteration of those GPUs will be air-cooled. It won't be liquid-cooled.
Yeah.
Just because they would've had to wait for the data center to be redesigned or retrofitted, and they're not gonna wait that long.
Right.
They'll think about the retrofit. Their retrofit will probably be on the first or second server refresh. I think 170 data centers go up to 150 watts. I mean, we're making sure we're ready, but at the same time, we have tremendous growth from lots of customers that are saying: I'm not gonna put the preponderance of my infrastructure at that power density, right?
Mm-hmm.
I'm gonna need a mix. I'm gonna need to retrofit a portion of my hall, so I can use a portion for GPUs, but the rest for CPUs. I think it's gonna be an amalgamation of both those types of infrastructure at the end of the day.
Yeah. So if you look across your portfolio of facilities, you know, what percentage of them are sort of, like, ready and able to accept significant power density, GPU-based workloads today? Which ones aren't? And then how do you think about adding things like other kinds of retrofits, like liquid cooling, et cetera, over time?
The majority of our data centers that you would retrofit are ready for the GPUs.
Mm-hmm.
And the minority that are not are legacy telco hotels that have thousands of cross connects, okay, in buildings that were never built to be a data center, that are not meant for AI or GPU workloads. They're gonna be the connectivity for the major metropolitan. These are landmark called internet gateways of 56 Marietta, 350 Cermak, Sovereign House, that just have a different use case than called AI compute. And we've been pruning our portfolio over time. We've sold out of billions of dollars of data centers that we didn't believe the long-term growth potential. Some of that was markets we didn't believe in, some of that was customer profiles, and some of that was infrastructure related.
So we think we're very well positioned for where this is going in terms of retrofit that customers may need on our campuses. Again, our campuses are multiple expansive buildings with some type of proximity, if not all, within a fenced area. Substations have the power infrastructure that can be already there or upgraded. So if and when we need to have a more rapid densification play, I think we're very well positioned.
Got it. And then from a geographical perspective, it seems like most of the activity has been concentrated in the U.S., but how do you think about the footprint in EMEA and APAC, and to what extent do you expect more AI to be happening in those places as well?
It's 100% accurate that we've kind of come full circle on maturity of data center markets, because the North American market was the most mature, most built out, the slowest relative growth rate versus outside the United States. But the AI first wave has certainly landed on the shores of the US, first and foremost, and I still think it's infancy to globalize. Part of that's likely due to the training workload. It's kind of going to the place where you can most efficiently and expeditiously build the infrastructure you need for the GPUs and the training. When I talk to customers, where I'm really getting my intel from, that I believe you're gonna see a globalizationist trend.
I don't think you're gonna see the size of the training replicate in these other markets, but it's interesting to unfold, which I think whether it's the United States or outside of the United States, is gonna be a much larger, call it, overall market and opportunity. I think the non-US markets will have a much greater share of that, and I think we're well positioned with these major campuses across Europe, Latin America, Africa, and APAC.
Mm-hmm. As you think about expanding your facilities, and also the differences in multiples between public and private assets out there, how are you kind of thinking about your appetite for acquiring data center assets from others, versus kind of doing new organic builds yourself?
I think the consolidation for the strategic puzzle pieces of the industry is kind of in the rearview mirror, not the front view mirror for Digital in particular, and I think our major rival as well. We've kind of picked up all the chessboard pieces, quite honestly, and we've been much more focused on organic market expansion, where there isn't a connectivity-oriented player to go buy, quite honestly. The lion's share of our capital spend has been development of new capacity in our traditional markets. We've entered a few markets in the last, call it, few years as well. I think that's much more of where we're putting our new investment dollars today, and I don't see that changing.
Mm-hmm. And, you know, you're well within, in fact, at the low end of your target leverage range, now. So how do you think about your appetite for further JV investments versus 100% ownership? And is there a certain kind of, like, IRR level you're targeting for the 100% ownerships, and what are you sort of targeting for JVs? How do you think about those two things?
Enterprise colo, interconnection-oriented opportunities, we especially due to the complexity, the attachment to our platform, for a host of reasons, we keep almost entirely on balance sheet, but one-off examples, like in South Africa with Teraco, opportunities like that. Hyperscale is where we pull the levers on private capital. We've done that on stabilized assets. We've done that now in development, and it's not about we'd like a project enough for our, we don't like, we like it enough with a partner, but we don't like it, or vice versa.
Right.
It's about, really, does it pass muster for us? We want to own it. We want to build it ourselves. We want to own it. And using those levers to essentially fund the business, that helps us to, call it, accelerate that bottom line growth, for next year, and then grow off that and compound thereafter.
Mm-hmm. You alluded to it when we were talking about power before, but you alluded to other factors that are constraining your ability to grow, whether it's nimbyism or state or local regulations or the rest of it. How are you thinking about, you know, the areas in which it's most attractive for you to build new facilities today? You know, balancing, you know, obviously the hot markets, where you could build all day long if you had the power and no constraints, versus the other ones.
So we are trying to make sure we're very thoughtful and good partner to all the constituents here, and make sure folks know that we're in this for the long haul. We're not in data centers or AI for a trade. We're trying to build long-term, appreciating, growing cash flows with long-term pricing power, and that means positioning in markets where we see runway for growth, where we're gonna be good neighbors, like next to the airport versus the housing project, things like that. We're giving back in terms of load shedding and giving back power when certain folks need it in different communities. We are looking at markets where we have different plays and potentially expanding our playbook.
In most of the major U.S. markets, we have what we call the connected campus. It's the series of the highly connected internet gateway, and one or two major campuses where customers can position their workloads and put their most latency-sensitive in the connected piece, and something that's a little bit bigger on the campus, along cloud compute. There are certain markets where we just have the connectivity angle, and we're thinking about expanding what we have in that market, but we've been doing business in that market, and we know it very well. So those are how we're thinking about the business, much more than we are not trotting out to this one-off market that a customer says, "Please build me a data center there," because we're gonna be there for the first renewal and the second renewal, right?
We're gonna be there, and we need to make sure that we're building long-term value.
Yeah. And outside of the U.S., how are we thinking about what markets are most attractive to build in? I mean, I'm assuming some kind of combination of where the customers have a presence, competition relatively weak, JV dollars want to go there, but how are you thinking about balancing that calculus?
I mean, we're looking at, I'd say, going back to playbooks that have worked for us successfully. In Europe, example, we have a real incredible position in Marseille, which is on the back of subsea cable, connectivity, now enterprise, and certainly a cloud computing market that blossomed out of nowhere for France. You look to the right and the left, Barcelona and Rome are examples, as well as what we've done in Greece. They may not become the Marseille, but we looked at those are. If we can call it, get the right locations, the right network density, the subsea cable, we're gonna build something of more lasting power there. Lasting value, not power in terms of energy.
In other markets where Frankfurt, we're looking at, where a market that's supply constrained, and we've got tremendous landholdings that we can potentially add on to on the periphery with adjacency, and make something big, really even bigger, and even more attractive as things, all the tenants' sites have been becoming, getting bigger in our category.
Mm-hmm.
Trying to exploit our advantages and continue to build a moat for our long-term business.
Yeah. You know, maybe just want to hit a couple of financial questions toward the end of this discussion, because obviously, that's what people ultimately care about to some extent. Clearly, occupancy rates have been fairly strong. You guided to one or two hundred basis points, increasing occupancy for this year. You've seen pretty strong new leasing activity. Do you feel like the environment we're in now, customers are actually pulling in leasing activity, such that, you know, they see a hot market, so they want to kind of get that allocation today? And how do you expect this sort of sustainability of re-leasing spreads to go over the next several quarters?
There's no question that, especially for the larger customers that have these AI workloads, that urgency is at the forefront. Fortunately, this is happening at a time when there's just broader supply constraints, right? So you can only pull in so much because the market only has so much. That's why a lot of these markets get to pretty darn full capacity. The enterprise colo business, I don't think that is experiencing the tightness from, like, AI demands. I think we're seeing there is that ourselves and another competitor are really pulling ahead, and there's not other attractive options that can deliver all this value to you as an enterprise colo customer. That's certainly advantageous to our economics. On the cash re-leasing spreads, we've just had this dramatic recovery in rates.
I don't know. I wouldn't. You probably would have asked me eighteen months ago, twenty-four months ago, would they have gone where they are today. I probably would've not expect them to run as far or as fast. I don't know whether they're gonna go another leg up or not just yet, but I know my expiration schedule is stepping down year over year, so I have an opportunity. Now, some of those contracts have advantageous rights to the customer. But the way things are moving with customers wanting to grow their infrastructure, densify their infrastructure, change their infrastructure, I think we're gonna have a good opportunity to either bring forward some of those renewals and capture that mark to market, even if the rate environment, the customer rate environment, stays where it is today.
Yeah. One question I get from investors often is just sort of like, you know, how you think about occupancy utilization, how that kind of factors into all this and the financials ultimately. You know, you report a portfolio wide occupancy rate of about 83%. Some markets, like Northern Virginia, in the nineties, others, like London, in the sixties. Can you maybe talk to why such a widespread in variance in that, if the market is kind of as constrained as we're talking about today?
I really think of the occupancy in two different buckets, called enterprise colo orientation and hyperscale. In the enterprise colo standpoint, you always have some type of kinda frictional churn or vacancy, but I still think we got a ways to go in that category, and I look at other examples that say that you could bring that up to closer to 90% over time. On the hyperscale side, it's obviously pushing up even higher than it's probably ever been, given the quantum of pre-leasing. And we're selling out of things that have been challenging. And lastly, on the margin, some of this demand for these newer upstart companies, smaller, are getting pushed to tougher to lease capacity because they're not first in queue, unfortunately, for the likes of our biggest customers.
Yeah.
That's a long-winded way of saying that we're actively taking measures on what we dispose of, what we monetize, how we're leasing, and how we're executing to continue to push that occupancy north.
Yeah. Including another kind of topic that I think is sort of central on a lot of, on investors' minds right now is sort of getting back to FFO per share growth, for the company, and what, run rate of growth you ultimately sustain at, over the next, you know, twelve, eighteen plus twenty-four months. How do you encourage investors kind of think about the algorithm for FFO per share growth?
We had to take a lot of steps backwards over the last several years to take big steps forward now. We understand that the driver of our value is accelerating, compounding, bottom line per share growth, right? That's why we may have been the only company that essentially gave two years out guidance on a guidance earnings call. We said called mid-single digits, and that's even muted because it's the path of deleveraging, selling out assets, joint venturing, which are long-term right plays, as well as other deleveraging, are still gonna be impacting us in 2024, 2025. What we said second is that's not the new go forward, that's the new floor, right?
And where we go from there is north of that mid-single digits, and the levers we're playing with then is how much development we do or how we fund that development, really.
Yeah.
So we're really trying to essentially continue to obviously put up the number we said last year for next year. Make that the floor, not the bogey, and then continue to grow off that. But also, at the same time, these levers we're pulling to make that runway of growth the longest possible, right?
Yeah.
Because these developments we're doing, these are good ROI projects. They're just raising your per share growth when they don't produce any income, and they're big, big, heavy capital projects.
Yeah.
Right? So, we have religion on this topic.
Yeah. Finally, just leave with everybody with one question, which is just broadly speaking, when you talk to investors, you've spoken with a lot of them over the last couple of weeks, I think. You know, what is the one thing that you think is misunderstood, if anything, about the Digital Realty story, and what's the one thing you would encourage investors to focus on?
So when I describe it, especially this, it's very apropos for this conference. Digital Realty, I view as a small but mighty boat in a big ocean of AI right now, that has not had the ups or the downs of some of our best customers and partners out there. And I know there's a lot of bulls and bears on the topic of AI. What I can tell you is this theme is additive to our business tremendously. From what we hear from our customers, it's not ending, it's still getting started. It's gonna be multifaceted, how I think we will support it, right? We've been almost only talking about big deals, big, large GPU training type thing.
But what, how customers use this, how enterprises engage with us, where they put that infrastructure, that's really a long tail of growth, I believe. And lastly, while we're in this so-called portion of the AI movie or whatever analogy you want to use, we are winning business that we're converting into long-term recurring revenue streams at attractive ROIs, the highest escalations, 3, 3.5, 4% contractual bumps, the longer durations, 15 years. And we're doing it in core markets. We're not betting on new markets that are gonna be there in 10 or 15 years. We're doing it in core markets, where even if we're wrong, there's other demand that is fomenting, be it cloud computing, enterprise IT. It's a different way, I believe, to benefit and support this long-term secular growth.
I think we're executing against it quite well.
Unfortunately, I've got 50 more questions and no more time. But that was a great conversation, so thanks, Andy, for being with us today. Appreciate it.
Thanks so much.