Good afternoon, everyone. My name is Brendan Lynch. I cover REITs here at Barclays. Very pleased to be joined by Arun Dev, Global Leader of Digital Interconnection Services at Equinix, and Katie Morgan from Investor Relations. Thank you for joining us.
Thank you for hosting us here.
It's my pleasure.
Thanks for having us. And just fun stuff before we get started. Today, what we're going to talk about is forward-looking in nature, so please check out our SEC filings for our full disclosures.
Great. So, Arun, why don't we start with an overview of your role at Equinix? Tell us where you fit in the ecosystem.
Sure. I lead our digital interconnection services at Equinix. I have responsibility for the overall profitability and growth of our interconnection portfolio. This includes our current products, leading new product innovation. I spend a large amount of time with customers, and when I talk about customers, it includes the likes of the hyperscalers, the clouds, enterprises, small and medium businesses, the telcos, and NSPs. What we hear from each of these customers really helps us shape our product portfolio, so as we continue to evolve our interconnection capabilities, my role is to really start to bridge what we're hearing from our customers, the problems that they're trying to solve, and providing the capabilities from a product perspective.
For those who aren't that familiar with data centers, maybe you could talk a little bit about what is the role of interconnection, what does that actually mean, and how does it differentiate Equinix's portfolio?
Great. I've got it in my title, so I can talk for a few hours, but I'll keep this a little brief here. So, interconnection is the ability for customers to connect securely, privately, with dynamically low latency within our data centers and across our data centers. Now, there's two flavors of interconnection. The first one is a physical interconnection. You've heard the term cross-connect. This is a point-to-point cable, so a customer can connect to another customer within our facility. Great for repeatable, large traffic flow use cases. It's simple, it's portable, and there's a number of different use cases it supports. The second flavor, Brendan, is virtual. Virtual interconnection is where, for dynamic workloads, you're able to bring up virtual connections. You have a lot more flexibility in doing this across data centers. You can go across metros.
So, you can set up connections across regions in the U.S. You could go to EMEA. So, the virtual option gives you a lot more flexibility for dynamic workloads.
Great. And we often hear about low latency being a key driver of demand at Equinix. Maybe you could talk about how the positioning of your portfolio can optimize for low latency, why it's particularly important?
Yeah. It's a great question. One of the best manifestations of the edge is our Equinix footprint, so we're in 73 markets across 34 countries, so if you look at our geographical footprint, we would have about 80% of the Americas, North America, Western Europe, and major population areas in Asia within 10 milliseconds off one of our data centers or locations. Now, location matters, especially as customers are getting more used to having things on demand, whether it's the app that they're using, whether it's the maps that they're using in their car. They want to be connected to the closest location with the lowest latency, and our global footprint and our reach allows us to be able to do that.
When you say 10 milliseconds, is that, in practice, what we're seeing? Is it that lower? There are hops along the way that might, in practice, push it a little bit higher?
So, typically, we see about 10 milliseconds within the metro. Now, when you go across metros, when you go across, that's when you start to see a little bit more of the hops. You start to see a little bit more latency. So, again, this is back to the edge, is that when in these major metro markets, you are within 10 milliseconds there. So, that starts to make a big difference from the experience that our customers are now able to give their users.
Great. Maybe we could talk a little bit about pricing for interconnections, how it's different by region, and maybe how it's different by virtual versus physical as well.
Sure. When we look at physical interconnection, the pricing is very simple. You're paying for the physical connection. So, it's a monthly fee that you pay. The customers are responsible for setting up the network equipment at both ends. Now, from a virtual, the pricing is a little bit more complex. It starts with getting a physical port. So, there's a monthly charge that you pay for your port. And the port comes in three flavors: one, 10, 100 gig. And so, there's a monthly charge for that. Now, beyond the physical port, virtual connections are on demand. So, as customers spin them up, there's a charge associated with it. Now, there's also a charge associated with going over regions. So, if you're running a workload in the New Jersey area and you want to get to Chicago, there's going to be a fee to go across to Chicago.
That's going to be different from the fee that that same workload now needs to reach over to London. We give a lot of flexibility with the virtual interconnection, but the pricing is a little bit more complex compared to physical cross-connects, where it's a simple monthly recurring charge.
Maybe we dig in on the physical side first. If I recall correctly, in the U.S., it's about $300 per cross-connect. Europe, maybe EUR 100. Asia, somewhere around $100. Maybe talk about the opportunity to potentially harmonize those price differences, the difference in price there.
Yeah. So, pricing in the U.S. is about $300. In Asia, it's about $200, and in Europe, it's about EUR 100. There is some regionality in there, right? So, for cross-connect pricing, just through the market dynamic, that comes into play. So, we don't see the three regions coming together with a global pricing, so there's some regionality associated with it. Now, from a pricing perspective as well, what we see with our customers is we want to be very conscious and thoughtful when it comes to pricing. And we really price the value, the value that we're able to provide. Now, we talked about interconnection. Interconnection is what makes our data centers some of that magic that happens in our data centers. When you're interconnecting at Equinix, you have access to a large ecosystem. We've been building this ecosystem for the last 26 years.
We've got 2,000 networks, 3,000 cloud IT service providers, and 10,000 customers. So, the interconnection part makes it very sticky for customers. They have access to all of these different destinations that they can get to at Equinix. So, we really look at, from a pricing perspective, what is the value we're able to give them. Specifically with cross-connects, in the immediate future, there's no plans, but in the medium to long term, we do see opportunities for enhancing pricing.
If you're pricing to value, what would you say is the difference that you're providing in the U.S. versus in some of these other markets where there is a lower price option available?
The key thing for us is the interconnection value we provide, the ecosystem of providers. Brendan, as I mentioned, we have 2,000 networks. The largest networks in the world are in our locations. Similarly, 3,000 IT cloud service providers are in our locations. So, when you look at it, North America has a really high density. In some of the newer regions like Asia, they're still developing, they're still coming up, and there's also price consciousness associated in some of the regions. So, some of that comes into play in what we can charge in those markets. Also, what does the ecosystem look like in each of those markets? Today, North America is the most developed ecosystem for us, for sure.
In terms of the virtual cross-connects options, you talked about different port capacities. How should we think about pricing for those different levels of capacity?
We charge based on the best way for me to describe this is like when you have tiers with your electricity or your water bill at home, and so the tiers are, it starts with the lowest end port is a one gig port, and that has a certain charge. There's some regionality associated with that as well, and then you have the 10 gig ports, which are very popular, and then 100 gig is the emerging trend, so in our Q3 earnings call, our CEO talked about strong demand in 100 gig, and so that's the biggest offering that we have, and then it's also, from a pricing perspective, we are going to charge for that bigger bandwidth that we support.
Okay. In terms of the stickiness of customers, maybe talk a little bit about how interconnection plays into that.
It plays a huge role, and then back to Brendan's earlier question around pricing as well: we price on value that we provide. So, it's not just space and power, but who can you connect to? Who can you connect to within the data center? Who can you connect to outside of our data centers? Which are the clouds that you can connect to? Which is the service provider that you can connect to? That's very sticky. The more interconnected the workloads are, customers get a lot more value from Equinix. What we see as well is when they first start to set up deployment, they know of a handful of locations that they want to connect to. As their workloads mature and they start to see who else is in the ecosystem, they have a lot more connections.
That unique relationship, we have about 116,000 unique relationships A to Z. It becomes very sticky, and then it reduces churn versus a customer who is running space and power, but is not really highly interconnected. We really focus on getting the workloads that are highly interconnected. Because there is value as well tomorrow, Brendan, if you decide you want to have your workload, people can connect to you as well. It's not just you connecting to others. We do focus a large part on ensuring that we have highly interconnected workloads.
The pace of interconnection growth has slowed a bit, certainly off a higher base. But back in 2019, it was like 9%. It's come down to about 4% this year. Maybe just talk about the dynamic behind that trend.
Yeah. I want to reference a comment that our CEO, Adaire Fox-Martin, made at the earnings, where she talked about the interconnection is back up like year over year, normalized constant currency basis at 10%. So, we start strong adds, 5,700 new interconnections. We have a total of 478,000 interconnections now at Equinix. Specifically, just going a little further as well, Adaire talked a lot about Equinix Fabric, our virtual interconnection offering, which is now a $250 million ARR business. Solid growth, driven a lot by the 100 gig ports, growth in 100 gig ports. So, we do see solid growth for our entire interconnection portfolio.
And maybe just to come back to the pricing element of those higher capacity ports, as you're selling, going from 10 to 100, is it a 10 multiple of the pricing, or?
I wish it was a 10 multiple. It's not. Our customers will be very upset if it's a 10 multiple. But no, it is a higher number. But what we're seeing is greater demand. And we kind of see this as a technology shift that's happening, is we are seeing fewer and fewer of the one. We're still seeing 10, but 100 is where it's taking off. And what we'll see is, over time, we'll see 400 and 800. I was at Gartner IOCS yesterday where I was getting questions. When are you thinking about 400? When are you thinking about 800? Things that are just very early in our evaluation.
And what would that take from Equinix's perspective to implement that? Is there technical things that need to be considered internally?
We do. So, we look at our global footprint. So, one of the advantages that you have when you come to Equinix is we have a globally consistent platform. And when we talk about, especially with virtual interconnection, is the virtual interconnection that you get in the U.S. is going to be the same that you get in Lima, Peru. It's the same one that you're going to get in Singapore. Pricing is going to be a little different, but the way that the system all comes together is globally consistent. And where this becomes important, Brendan, from a technology perspective is when we look to upgrade, we will start to offer these in these markets in a consistent manner. Now, we'll stagger the rollout. We're going to go where the demand is. Now, I would say, if I look at 100 gig ports, North America is very strong.
Europe is strong. Asia is emerging. But we'll have a rollout plan, and we look at by market, but we will ensure it's globally consistent.
If we go back a few years and look at the cloud offerings, they were primarily focused on compute and storage. It does seem like the cloud is getting more into networking, which was historically kind of the bread and butter of what Equinix was offering. How should we think about that evolution there?
The way that we look at the market is the areas that we're extremely strong, where we build our products and solutions around colocation and interconnection. And so, specifically around interconnection, we recently launched Equinix Fabric Cloud Router. It's an addition set of capabilities to Equinix Fabric. And what Cloud Router does is it allows you to connect to clouds. It's a virtual routing service connected to clouds, so you can connect and between clouds. So, what this capability offers now is if you have a workload, whether it's at Equinix, whether it's in the cloud, you now have the ability to use a virtual service to be able to connect between the clouds and between your colocation deployments as well. So, we're continuing to invest in innovation around areas where we feel we have a right to play and a right to win.
Specifically, as you see with Cloud Router, it does to the cloud and between the clouds. Now, some of the benefits that come with Cloud Router is it's a virtual service, similar to what you would get in the cloud. So, there's no hardware, there's no maintenance, there's no licensing to deal with. It's as a service. You bring it up, you use it. If you need to bring it down, you bring it down. You just pay for what you use.
In terms of the power demand for new networking services like virtual Cloud Router, can you talk about how that has evolved, how that power density has increased?
What we've seen with just customer behavior is the clouds have trained them really well in what to expect when it comes to networking. It's as easy as possible, as simple as possible. It's delivered as a service. So, when I sit with customers, one of the biggest requests I get is, can you give me a similar experience? Can you make it as easy for me to run this where I don't have to deal with side vendors? I don't have to, I don't need licensing. And so, from our perspective, offering more of these capabilities like Cloud Router is really solving real customer pain points in an easier way. Now, from our, if you look at it from a power perspective, deployment perspective, it's really built for us in software. It's built globally consistent. The networking nodes itself don't take a lot of power.
But it's really the form factor that we expose to our customers. And we're trying to simplify it, make it more cloud-like. And what we're finding is, as soon as you offer something, they come with more. They want more. So, they're hungry for more of that. And part of that's driven by the fact that skills are hard to find these days. Especially networking skills are hard to find. The architectures are getting more complex. What we hear from our customers is they're all multi-cloud. Each cloud operates a little differently. They've still got on-prem, they've got branch offices, and they don't have the skills. They don't have the appetite as well to go invest in physical hardware at times and set up all of these capabilities. Rather, what they're asking for is, give me a cloud-like experience. I need to connect to the clouds and between the clouds.
That's kind of where Fabric and Fabric Cloud Router shine.
It does seem like a very good opportunity to tap into the multi-hybrid cloud environment. I do wonder if there's a risk that some of the networking workloads that Equinix would have had in the past are now just kind of concentrated in the clouds themselves. So, maybe enterprises in the past would have gone through a similar offering in the past at Equinix, but now they can do it internally in the cloud. Do you see that as a risk?
I don't see it as a risk. So, Brendan, as I mentioned, I spent a fair bit of time with customers in all three regions. I've only had one customer in France tell me they were mono cloud. That was a new term for me, mono cloud. Every other customer that I've talked to is for various reasons multi-cloud, whether it's an acquisition, whether they were shadow IT, they're all multi-cloud. And so, what we hear from them is they've also got branch offices. They've got corporate offices. And now, while most of their workloads may be in a specific cloud, they still have other workloads in other clouds. They've got these branch offices, other premises, and they need to be able to stitch all of this together. And this is kind of where they come back to Equinix, like having the network nodes at Equinix.
It could be we support different flavors. You could bring your network equipment, run it at colocation, or you can use services like Fabric and Fabric Cloud Router.
I think you now have, speaking of that, 200 cloud on-ramps. How do you anticipate that's going to grow over the next couple of years?
We have the market leadership position in cloud on-ramps. I think we have 2x the number of cloud on-ramps compared to competitors. I think about 40% share of the private on-ramps. In most of the metros, the major metros where we see customer demand, we do have on-ramps. And the on-ramps are an important part of our ecosystem. So, it's really how you get to the clouds. We've had conversations earlier today as well as people do want to get to the clouds. The workloads are in the clouds. The users may be elsewhere. Their branch offices may be elsewhere, but they have to be able to get to the clouds. So, we are very conscious of these areas where we see new markets we may be entering or new markets developing where there are clouds and requiring on-ramps.
So, as I mentioned, we've got 40% of the private cloud on-ramps, and we continue to evaluate in some of the new markets like Jakarta that we're entering, Chennai and others, adding on-ramps.
Great. Let's switch gears a little bit and talk about inferencing. I'm sure it has a lot to do with everything.
I was waiting for the AI question.
As AI shifts from training to inferencing, where do you see Equinix playing in that space?
I want to go back to what I shared at the beginning. We're in 73 markets in 34 countries. The location, the global reach that we have, the neutrality, the ecosystem we have plays a huge part when it comes to inferencing. When we look at these locations that we're in, 80% of the population, the Americas, Western Europe, and major metros in Asia are within 10 milliseconds, so from an inferencing perspective, you are within that region where your experience is going to be great, and from a customer standpoint as well, what they observe is there are specific markets that they want to serve. Being at an Equinix facility, you have on-ramps to the clouds. You have close proximity to the end users that they're targeting. We're able to provide that.
To what extent are you seeing AI-related demand already?
We're very early in the AI journey in our customer conversations. But from an Equinix perspective, it's a multifaceted approach that we're taking. One, I want to start with the large hyperscale demand. Our xScale business, where we build the larger data footprint data centers, is a direct beneficiary of this appetite for large hyperscale AI demand. We also have enterprises that are coming to Equinix to where they want to be able to run their AI workloads at Equinix. Our data centers are AI-ready. We also have great partnerships with the likes of NVIDIA, HP, Dell, WWT. In addition to this, we also have the GPU as a service providers that are at Equinix. They run their network nodes. We've talked about in previous earnings calls about with Lambda Labs, CoreWeave. Pair all of this with our global reach. You're in 34 metro markets.
You do have the right set of capabilities for these AI inferencing. But what we're also hearing is, right now, they're looking for early prototypes of value. We see this really pick up in the medium term where the true inferencing there, because everybody's still trying to figure out what to do with AI.
Sure. Maybe you could talk a little bit about the power requirements for AI training versus what would mostly be AI inferencing within Equinix facilities.
From a training perspective, if you look at what we typically see with customers, is they have private AI deployments. So, one is they've got higher density. They have required things like liquid cooling. We have over 100 facilities that can support liquid cooling. But it still requires the customer to set up the infrastructure, manage it, and run it. Now, with the scarcity of GPUs, this is not an easy task for them. So, we have a different set of requirements from a power perspective for somebody that's running all of their training at Equinix. From an inferencing perspective, it's much smaller because the training's being done elsewhere. It could be done in the clouds. It could be done at Equinix. But the inferencing is done from the edges. So, the power requirements are significantly lower for inferencing than for training.
Training is really driven by a couple of things. One is the availability of GPUs. Second is data sovereignty. Third one is data, I would say, security where we have enterprises tell us they do not trust a third party with their data. And so, those are kind of the factors that come into the training aspect.
I would think that it would be very expensive to train at Equinix as a premier product offering. It's just generally going to be more expensive than some other data centers that are out there. Can you talk about the nature of the training workloads that you are actually seeing?
Yeah. What we would see, just one of the examples that comes to mind is in the pharmaceutical industry. So, large training, their data is proprietary. It needs to stay within their control. So, that's what we allow. Our facilities allow them full control of their data. It's not now being transferred elsewhere. And it's interconnected. So, if they need applications that are running in the cloud that need to get to this data, we provide the ability for them to do that. Pairs this, Brendan, with our power density that we can support, the liquid cooling that we have. It's attractive for these use cases for either for data, as I said, data security or data sovereignty where they will run their training at Equinix.
Great. Let's talk a little bit about edge opportunities. This got a lot of attention a few years ago. I think Equinix looked pretty wise for not going into it too aggressively because a lot of it has not really materialized to the extent that it was anticipated just a couple of years ago. How do you see the edge opportunities going forward?
So, when we talk about edge, and I want to go back to an earlier statement that I made, is we believe Equinix is the best manifestation of the edge. It goes back to our global reach. The fact that we're in 73 markets gives us an advantage in where we're now able to serve a large portion of the population that we're looking to reach. So, from an edge investment perspective, we've been very specific about these markets that we've entered. We also have the interconnection density and the neutrality. So, all of this comes in, works in our favor. Now, the far edge is still very early days. So, we're taking a wait-and-watch approach in where the far edge goes. We believe the current footprint that we have, we are able to cover most of the edge opportunities, the workloads that we're seeing within these locations.
Maybe just to tie that also into inferencing, to the extent that there is some inferencing at the edge, being the edge device, the cell phone or the vehicle, how does Equinix fit into that scenario?
Here's a common example that you hear about. We take automotive as an example. If you look at self-driving, a lot of that is trained at a central location, and the self-driving gets better over time, and all of that training happens at a central location. Now, from an actual inferencing perspective, when you talk about self-driving, it happens on the vehicle itself. On that vehicle, you could ask for what the weather report is, where the nearest Shake Shack is, and that inferencing happens more at the edge. You can ask it a set of questions. That inferencing happens at the edge. It's not an immediate, like, if I don't brake, I'm going to run into the car in front. It's more of, I'm asking it. That level of inferencing happens at the edge.
So, that's how we see examples of what inferencing has been used for right now, which is they're still within that 10 millisecond latency. You're still getting answers back. But for critical ones that need to be real time, it happens on the device itself.
Maybe just one last question looking forward. Where do you see the most space for innovation at Equinix? What gets you the most excited?
As I mentioned, we spend a lot of time with our customers, and I think what we hear from them is the two areas that we're extremely strong at is colocation and interconnection, and solutions, products that make it easy, low friction to consume these services is what they're looking for, so from a product innovation perspective, that's really where our focus is. When I talked about cloud router, it really made routing easy for our customers. It made it simple as a service, and so they're looking for more of those capabilities, so we're going to continue to focus more on the colocation interconnection space. Now, besides this, the other piece that we're working on with all the service providers is more API-driven integration. We have, as I mentioned, there's 2,000 networks, and some of the networks are more advanced than others.
But we're trying to make it easier for new end-to-end integration. So, at the end of the day, what we want to provide our customers is solutions that are simple and easy to use around colocation interconnection.
Everybody appreciates simple and easy.
Yes.
Great. Thank you very much, Katie. Thank you.
Thank you Brendan.
Thanks.