Google VP Majd Bakar speaking at the Game Developers Conference in San Francisco in March 2019 JOSH EDELSON/AFP/Getty Images
Google VP Majd Bakar speaking at the Game Developers Conference in San Francisco in March 2019

How Cloud Giants (Hyperscalers) Go About Leasing Data Centers

Data Center Podcast: Stack Infrastructure’s Tim Hughes on the business of hyperscale computing facilities.

Tim Hughes got his start in the data center industry when his friend offered him a contract position at Facebook’s first data center. The social network was leasing space at a DuPont Fabros Technology (now part of Digital Realty) facility in Ashburn, Virginia. The job wasn’t glorious. It entailed pulling pallets, sweeping floors, racking and stacking.

“Everybody that has worked inside of a data center knows, that’s some of the manual labor associated with building the internet,” he said.

He rose through the ranks and eventually became part of Facebook’s site selection team. By that point, in 2014, Facebook had been building its own data centers in addition to leasing them, designing and operating some of the world’s largest and most advanced computing facilities.

Because he started at the very bottom of the organizational ladder and because the organization was relatively small and cross-functional, he got to learn every aspect of hyperscale infrastructure planning and operations, Hughes told us in an interview for The Data Center Podcast.

His current role is director of strategy and development at Stack Infrastructure, which was formed two years ago with a simple thesis that data center growth will continue and that most of it will take place in the realm of hyperscale cloud platforms. Stack’s strategy is to take advantage of that growth by building and leasing to hyperscalers the part of their capacity that they, for one reason or another, will not build on their own.

On the latest episode of The Data Center Podcast, Hughes took us inside the world of hyperscalers’ data center strategy. We talked about how operators of the world’s largest cloud platforms decide where and when to lease data center capacity versus building their own, how the influx of new capital has changed the wholesale data center provider business, and how Stack differentiates itself in this extremely competitive market.

Have a listen below or any other place where podcasts live (Apple Podcasts, Spotify, Google Podcasts, Stitcher):

Tim Hughes, Stack Infrastructure, on The Data Center Podcast

FULL TRANSCRIPT:

Yevgeniy Sverdlik, DCK:

Hey, everybody. Welcome to the Data Center Podcast. I am Yevgeniy Sverdlik, editor-in-chief at Data Center Knowledge. We have today with us, Tim Hughes, he is director of strategy and development at Stack Infrastructure. They're a wholesale data center provider. Tim, you're in Denver. Right? So, tell us, what's it like in Denver nowadays? Is it cold? Is it quiet because of the pandemic?

Tim Hughes, Stack:

Yeah. I mean, it's pretty much like everywhere else, quiet because of the pandemic. But we just got hit by the zero degree weather over the weekend, that I think found its way to a lot of the rest of the country. So, we got to experience that over the weekend. And then I went outside and it was 25 degrees out. And I was like, "Man, this feels good. I should just put a t-shirt on." The relative difference between negative five and 25, you realize it's a 30 degree swing, it's exciting stuff.

Yevgeniy Sverdlik, DCK:

Well, and you have electricity, which is always nice.

Tim Hughes, Stack:

Totally. That's an interesting case study that's going to come out of that with how ERCOT manages their grid because they do a lot of really good stuff, but clearly they missed the mark on a couple of things.

Yevgeniy Sverdlik, DCK:

And I think you're my first podcast guest that has actual podcast recording equipment.

Tim Hughes, Stack:

Oh, that's good.

Yevgeniy Sverdlik, DCK:

You used to have your own podcast, right? What was that about?

Tim Hughes, Stack:

Yeah, that was 10 years ago. And so, I can be forgiven, I hope for my attempt at comedy podcast. I still think it was funny. But yeah, a buddy of mine sat in my basement and recorded, I don't know, a hundred hours worth of podcast content. And the easy part is recording podcasts. The hard part is editing. And so, I'm going to do my best to make a clean edit for you, so we don't have to go back and fix a bunch of stuff. So, I'll make sure I mute all of my coffee drinking.

Yevgeniy Sverdlik, DCK:

Here comes the coffee drinking.

Tim Hughes, Stack:

So, we're good.

Yevgeniy Sverdlik, DCK:

Thank you. I appreciate it. So, we're going to talk about Stack. We're going to talk about wholesale data centers. We're going to talk about hyperscale. But I want to talk about your, I guess not previous job, but your past job, you spent many years working at Facebook, doing all things, data center there. How did you end up at Facebook?

Tim Hughes, Stack:

Yeah. I started at Facebook in 2008. I think it was November 5th. There was a presidential election the day before, so stuck in my mind. Yeah, so I started there pulling pallets and sweeping floors, and racking and stacking. Anybody that has worked inside of the data center knows that's some of the manual labor associated with building the internet. And a lot of the work that we were doing then is actually done offsite in integration facilities, now is as these facilities continue to scale up.

Tim Hughes, Stack:

But I got brought in from a buddy of mine who realized that I was working in a back room, in a retail establishment at Target actually overnights. And he was like, "You know what? Data centers are like this. You have to go around and manage things, organize things, deploy things. There's just a lot more software involved." And so, he reached out to me, gave me the opportunity to start as a contractor in Facebook's first data center on the East Coast in a DuPont Fabros facility, actually ACC4. And worked my way up from there through racking and stacking, to configuring switches and provisioning servers. And I eventually became a program manager and deployed Facebook data centers and their Facebook edge network, and dark fiber facilities. And found my way onto the site selection team for Facebook and did their data centers in Denmark, in Ireland, and managed the leasing portfolio for a year, year and a half, before heading out and taking a mini sabbatical, and then starting my own consulting company, before joining up with Stack about a year and a half ago.

Yevgeniy Sverdlik, DCK:

Well, so from racking and stacking, from sweeping the floors to the upper echelons for Facebook infrastructure strategy.

Tim Hughes, Stack:

Yeah. The middle echelons, I think is probably the right way of putting it, but I got to work on a lot of really fun and big projects, which I'm really fortunate to. And you only get, I think a couple of, if you're fortunate, you get one opportunity to do that in your career, where you're starting at a small company and you're actually able to influence large decisions and you have enough flexibility to actually push things in certain directions at a company that's growing quickly. I'm fortunate enough to have now two with Stack Infrastructure here, but that was a lot of fun.

Tim Hughes, Stack:

And to your point, there's nothing you can't really discount starting really at the bottom, at the absolute starting point of where you can learn everything that you can in an industry. And I was able to do that with data centers.

Yevgeniy Sverdlik, DCK:

Right. Because then, once you do move up the ranks, you know through and through everything that's going on. So, when you manage teams, you know what kind of things they face day to day, that must make things a lot easier.

Tim Hughes, Stack:

I think you can empathize with people as well too, because you say to yourself, "Okay, we're going to site a data center in this really great location from a power cost perspective or from a tax incentives perspective, but it's in the middle of nowhere and that doesn't have a quality of life for people that would be living there." Or when you're making design decisions on facilities and you ask yourself, okay, very simple things like how do we ensure that if you're in a facility like what we build, how do we ensure that our clients have the best experience when they're coming in the door or receiving a package, or taking the trash out? These are like very minor seeming things when you're in the grand scheme of designing and are really important for the people that work there every single day. And so, you get to think from that perspective.

Yevgeniy Sverdlik, DCK:

Because you've been that person doing the shipping and receiving.

Tim Hughes, Stack:

I took a lot of trash out, so I know it's nice to have the box trashcan closer to the loading dock, so you don't have to walk in the weather that we're experiencing on the East Coast. Box does still needed to go to the recycling bin whether it's five degrees or 95 degrees out.

Yevgeniy Sverdlik, DCK:

Yeah. The fine details of the user experience. And so, you went from site operations to site selection. Can you maybe tell us that story? How did you end up on the site selection team?

Tim Hughes, Stack:

Yeah. So, I took an intermediary path through the technical program management team within the infrastructure organization at Facebook, which essentially is at the time, I believe still is to this day, like one of the most cross-functional teams that you could deal with from a technology perspective and infrastructure. I think that's true in a lot of these organizations, large tech organizations. And so, there was a very small team of us building a lot of infrastructure quickly. And you have to touch a lot of different pieces, everything from ensuring that you're deploying servers in edge nodes around the world, to working with very smart and talented software and production engineers to build the software stacks that support these things.

Tim Hughes, Stack:

And oftentimes you're doing stuff at a scale that hasn't been done before. And so, both from a physical infrastructure perspective and a software perspective, you're learning how those pieces fit together, serving a billion plus users that maybe others haven't done before.

Tim Hughes, Stack:

And so, I had a lot of insight into the cross-functional nature of the infrastructure organization. And so, the site selection team, my current boss, who's the chief strategy officer at Stack Infrastructure, a guy named Matt VanderZanden, he reached out to me and said, "Hey, we want to increase our cross-functional engagement with the rest of infrastructure and data center site selection." Because they felt like they were maybe getting a little bit too siloed. And they wanted to say, "Hey, we want to make sure that we're working with and collaborating with all stakeholders from a network perspective, from a software perspective, et cetera."

Tim Hughes, Stack:

And so, he brought me on and I said, "Hey, look, this is something that I'm very passionate about. I'm very interested in deploying the largest scale pieces of physical infrastructure that Facebook or other tech companies are working on." And I've always had a passion for acquisition and negotiation. And so, I was able to get my hands on that and I was able to provide some kind of cross-functional support to the team as well.

Yevgeniy Sverdlik, DCK:

That's interesting that you brought up cross functionality and how these hyperscalers, I think they really value that. Right? And that opens up opportunities when you're working in an organization like that when it's not so siloed and segregated, and there isn't a rigid hierarchy. There's a lot of opportunity for people to move from one team to another, and then just even see what, if you see what other parts of the company are doing.

Tim Hughes, Stack:

Yeah. And at some point, you get so big that you can't really do that, but I was able to grow in lock step with the organization while they grew. And I think a similar thing is happening at Stack where we're, I think we just passed a hundred employee milestone a week ago or something like that. And so,, we're seeing the same thing here, right? We're building this organization. And we're growing really quickly. And so, my goal is to grow and scale myself and our strategy and development organization in the exact same way.

Tim Hughes, Stack:

And the interesting thing is, as these organizations get larger, there's just natural additional friction or inefficiencies as they grow larger. It's just like the multi-talker problem where you have, instead of communicating one-on-one or one to two, you have n times, n minus one types of communication, where your stakeholder group grows and grows.

Tim Hughes, Stack:

And so, if you're in those internal organizations, some of the best things that you can have is good external partners, like a Stack Infrastructure that can be fast, that can be nimble. We're still small enough to make those quick decisions and respond to our clients. And that's certainly a little bit of a plug for our organization, but also just a reality, having been on the other side of the table, that you need that outlet of those nimble fast partners that maybe are three to six months ahead of where you are to be able to grow your organization and quickly, even if you do the majority of your data center capacity self-perform.

Yevgeniy Sverdlik, DCK:

You can outsource the nimbleness element if you are so large already that maybe you cannot be as nimble. That's interesting.

Tim Hughes, Stack:

Absolutely.

Yevgeniy Sverdlik, DCK:

So, were you at Facebook when they started building their first owned data center, the one in Prineville, Oregon.

Tim Hughes, Stack:

Yeah. That effort started in late 2007, early 2008. And I joined at the end of 2008. So, they had started some of the site selection efforts prior to me being there. And look, I was just a guy racking server, so I'm not going to take any credit for that. I know the people that worked on it, very smart people, some of them that are still there. But I was there during the initial turn-up of the facility, working with our provisioning teams and our software engineering teams to bring that facility online as a new region. And it created interesting problems from a software stack perspective because we were at that point just to regions in East Coast and the West Coast region, in Santa Clara and in Ashburn, and then they added this Prineville region.

Tim Hughes, Stack:

And so, the question was, how do you best optimize the software stack for a new region that's also on the West Coast, but far enough away that if it's not its own segregated region, then you would create latency issues? So, that was a lot of interesting problems that other people were working on that I got to watch, while I racked up a big Cisco and Juniper switches and routers, and helped to provision servers.

Yevgeniy Sverdlik, DCK:

And the Stack Infrastructure, you mentioned not an old company. It was very recently formed in 2019. And it was formed by a combination of some T5 assets and the Infomart assets that were left after Equinix bought Infomart Dallas. It was formed to go specifically, was one of the few companies that were formed to go specifically after the hyperscale data center outsourcing opportunity.

Yevgeniy Sverdlik, DCK:

Now, the interesting bit about Stack is, one of its Becker's ICONIQ Capital, I'm not sure if they're still involved, but they were in the beginning. And they're a wealth manager for some very rich Silicon Valley figures, people like Mark Zuckerberg, Sheryl Sandberg, Jack Dorsey. Can you maybe flesh out the business strategy a little bit? And I mean, I summarized it, going after the hyperscale opportunity. What's the strategy? What are Stack's differentiators?

Tim Hughes, Stack:

Yeah, sure. So, while we were putting the company together and I started a year and a half ago, so I wasn't there at the formation, but I've been able to continue to work with the folks that were. The initial hypothesis is that simplistically, but importantly, data center growth is going to continue in the near medium and long-term. And that growth is going to happen primarily at the larger scale side of the market, to your point, the hyperscale side. And so, we should develop a company that's specifically organized to capture that.

Tim Hughes, Stack:

So, the strategy was relatively straight forward, where we worked to develop a national platform and saying, okay, instead of just starting with capital and dirt, which really anybody can do, we wanted to create a cohesive national platform first on the back of performing assets to your point, that were acquired through T5 or Infomart. And those assets have served us extremely well to create this revenue base for the company, as well as the ability to show that we can own and operate data centers.

Tim Hughes, Stack:

There's a lot of capital right now chasing data center development, as you well know. And anybody can say, "Hey, I have a hundred acres and I have a lot of money. But we were able to very quickly show that we had the organization and the talent to actually operate these data centers as well. Because nobody wants just the data center, they want somebody that can operate the data center well, and we're doing that.

Tim Hughes, Stack:

And so, once we establish that platform, our next step was to begin to develop new facilities. And so, we did that in Chicago, Silicon Valley, Portland. We have some Northern Virginia projects going on as well. And so, we're quickly deploying additional capital and building additional data centers for the clients that we now have very close relationships with from all of the hyperscale players.

Tim Hughes, Stack:

So, it was a key component of ours to be able to prove that competency through operations, then to prove our ability to develop these facilities in a very Stack specialized design, but not so bespoke that if a client came in and looked at it, they would be like, "What is this? I don't understand this mechanical system. I don't understand this electrical system." And in a way that's modular enough for folks to be able to make changes if they need to, and for us to scale up rapidly and from a cost-effective perspective.

Tim Hughes, Stack:

So, that's the strategy, create the platform, establish operational competency, begin to develop our own projects in a way that with a product that our clients are interested in. And then really develop that deep relationship with those clients. And create that sense of us being the outsource, nimble partner that you're looking for. And hopefully, we're in markets that we're ahead of, even our clients from a schedule perspective that we can quickly perform.

Yevgeniy Sverdlik, DCK:

As you mentioned, there aren't that many experienced teams that aren't already working somewhere building data centers and leasing them to big clients in 2019. And now you guys are up some serious competition. It's a very competitive market with some very experienced companies with long track record, like Digital Realty's of the world, and CyrusOne, QTS. When you're talking with the Microsoft or with the Facebook, how do you explain to them, why Stack and not Digital Realty?

Tim Hughes, Stack:

I can't give any client names specifics just because of NDAs, as I'm sure you're well aware.

Yevgeniy Sverdlik, DCK:

Those are just examples.

Tim Hughes, Stack:

Of course, yeah. And so, look, I think we have the best team in the industry right now, and that's not trying to throw any shade at any of our competitors. And you would say, "Yeah, of course, naturally, Tim, you would say that because you work there." I wouldn't have joined this organization if I didn't think that we had some of the best folks in place.

Tim Hughes, Stack:

And really there's been a shift in the last three or four years from the data center industry being a place in which it was very real estate focused to a place in which it's very much more driven by the finance side of things. You see different types of capital entering into the data center space. And that means that different players, individuals that are contributing to this space are entering it as well, including a lot of the people that I work with. And we have people that have been in the industry for 20 years. And we have, I think are probably average time in the industry is 10 to 15 years.

Tim Hughes, Stack:

But in terms of the management and in terms of what these companies are expecting, the competition for the Digitals and the Cyrus is completely different from when they were developed, right? And the problem sets are completely different from when they were started as companies as well. Not to pick on anybody in particular. But we are developing ourselves specifically for the problems of today. We don't have 20 years worth of legacy assets that we have to manage and think through. We are able to remain nimble and small and focus on the core customers that are at the top of our list, and we're at the top of their list, we think too. And then we're able to hire really good people because we can keep the team smaller.

Tim Hughes, Stack:

And so, I think we have some of the best sales organizations, some of the best finance organization, some of the best engineering and design people in the industry right now. And a lot of our folks come from the hyperscalers themselves. My background, my boss's background, I would say probably a good 50% of our organization have some sort of hyperscale background from one of the top four or five companies. I think that that's pretty unique. So, we're not just coming from a real estate background, or we're not just hopping over from somebody else directly in the industry. We know what our customers and our clients are looking for.

Yevgeniy Sverdlik, DCK:

So, we keep talking about hyperscale data centers. That word means a lot of different things to a lot of different people. Some people look at it from a size perspective. Some people look at it from the application that these data centers enable. Some people look at it from a design perspective, there is a certain way hyperscale data centers are designed. What's a hyperscale data center to you? How do you think about these things?

Tim Hughes, Stack:

I just look at it as scale. I mean, it's hard to look at it from an application perspective because many of these hyperscalers are cloud providers. And so, what applications are they're running? They're running whatever applications their clients want to run. And then if you look at it from how it's being designed, I mean, there's a hundred thousand different ways that you can design a data. There's infinite ways you can design a data center. And then, there's probably like 10 that are actually good, right? In terms of what's out in the market right now. But that's going to change over the next five to 10 years. I mean, we have a bunch of different trends in electrical topology design. You have a bunch of different trends in mechanical design. And those are all different from 10 or 15 years ago.

Tim Hughes, Stack:

And so, really for me, it's just the scale at which you're deploying and/or buying. And so, that's why we view a lot of our clients as hyperscalers, is because they're self-performing at very large scale. And they're working with us to develop projects at very large scale as well. And so for me, it's mostly just, pick an arbitrary line, I'm sure there's a way you could define it, call it 12 megawatts, 24 megawatts, 50 megawatts, a hundred megawatts. That's more of how I think about it, is in terms of the raw capacity of the facility. Not raw capacity, I would say gross capacity after P, we have the facility.

Yevgeniy Sverdlik, DCK:

So, five megawatts in op or something like that.

Tim Hughes, Stack:

Yeah. I mean, that's pretty small still, honestly, these days.

Yevgeniy Sverdlik, DCK:

That's too small for some.

Tim Hughes, Stack:

Yeah. I mean, I think hyperscale is the place at which you can develop a facility and capture the economies of scale to be competitive in the market for what a hyperscale buyer is looking for. You can certainly build smaller facilities and they make perfect sense for the air "retail side of things" in which you have 2N plus one redundancy because it's network equipment or it's edge nodes, and there's load balancers in there. And you can't lose a rack, otherwise you lose your entire application or website. And so, those things certainly exist for a very good reason, and they will continue to exist and continue to grow. But what we're doing is more of the core application data center side of the world, right? And that's a place in which you have to deploy at scale to capture the economies of scale, to be able to be competitive in the marketplace, whether it's against others that are in the same business model as you, or against the types of projects that your clients themselves with self perform on.

Yevgeniy Sverdlik, DCK:

So, 10 megawatts, what's the minimum [inaudible 00:23:03], cut-off for you?

Tim Hughes, Stack:

Honestly, I would say 20 plus.

Yevgeniy Sverdlik, DCK:

20 plus.

Tim Hughes, Stack:

In terms of the facility size. And that you start to capture economies of scale there both from a construction perspective, which is very important. And there's a number of inputs that go into construction, everything from the procurement of the raw materials, to the earthworks that need to be done to your supply chain of critical components. And we have a very well-developed supply chain in which we spent the last year putting that together, where we worked directly with our supplier partners, everything from generators to our air-cooled chillers, to our PDUs, to our fan walls. And we worked with them to develop a scale relationship that provides us with economies of scale from a pricing perspective. But also, I think even more importantly, the ability to work with them to refine a delivery schedule through that supply chain, that actually makes us more competitive from a schedule perspective for delivery than others in the industry, as well as self-performed from our clients.

Yevgeniy Sverdlik, DCK:

So, we've been tracking this explosion of data center capacity for 10 years now at DCK about that. It's accelerated in the last five or so. Where in the life cycle of explosion are we today, do you think? Are we are at the middle, the end or still the beginning?

Tim Hughes, Stack:

The interesting thing about data is, it's not constrained by any clear physical limitation. I think if you look at other industries, oil and gas is a really easy one. I mean, you can always say, "Hey, there's going to be peak oil at some point. There's just so much." Maybe that's in 10 years, maybe that's in a hundred years. Maybe we've already hit it. From a data perspective, whether it's processing, whether it's storage, there's not actually a clear thing to point out. So, it's really hard to say where we are in that life cycle.

Tim Hughes, Stack:

And I think as we continue to trend towards evermore data-driven applications that continue to improve our lives most of the time, sometimes distract us with cat videos, it's not as improving. That's, I think we're going to continue to have a demand for data services for data centers. And I think you're going to see it in different types of flavors, but the current trend is certainly towards this centralization of servers in the cloud. And as folks centralize out of, I think there's some studies that say we're about 50% through that cloud transition.

Tim Hughes, Stack:

So, if you're looking for a specific epic in this ongoing life cycle, you can point at the cloud transition and say, "Okay, maybe we're about 50% of the way through the cloud transition." And the reason that that's important is folks are transitioning their IT services out of closets, out of small data center deployments and into the cloud. And so, at least from our perspective on the hyperscale end of the business, we see that continued movement into centralized cloud locations as a continued growth in the hyperscale side of the industry. So, that's say, that's 50% through from from a cloud aggregation perspective.

Tim Hughes, Stack:

But the other things that we're seeing that are not quite as easily tracked and confined, and it's really hard to predict where this goes and where this ends is, we continue to see increased client traffic. And clients importantly, aren't just people, but more and more people are getting on the internet. I mean, there's some studies that show we're probably going to be about two thirds internet penetration globally in 2023, three years. That's about 5.3 billion people. Three years ago, we were at 3.9 billion people. And so, that's a pretty significant increase just in human eyeballs.

Tim Hughes, Stack:

And then, people use the buzzword IOT all the time, internet of things, but really that is an increase in clients as non-human interactions increase on the internet, as our smart devices, as our cars, et cetera connect. That's as heavy from a data center perspective as human beings get on. So, those are two significant trends that it's really hard to predict, when that growth cycle is going to stop from a, just a new client perspective when you add together both people getting online and devices.

Yevgeniy Sverdlik, DCK:

Okay. So, the answer is basically we don't know, but things are still, it's still growing pretty quickly.

Tim Hughes, Stack:

10 years, we'll continue to have significant growth through this cycle. As more people come online, as more devices come online, as cloud aggregation continues, as the complexity of the applications that we're online with continue to increase, and as people spend more time online, the next 10 years, we'll certainly see continued growth. Beyond 10 years, I mean, it's hard to forecast anything, right? But I would expect it to continue to grow. And the question is, are we seeing the same types of growth trends that we've been seeing over the last five years? Or is it more of a stabilized mature industry?

Yevgeniy Sverdlik, DCK:

Or is there more growth in some places around the world and less in others where a lot of capacities have already been built, right?

Tim Hughes, Stack:

Certainly. And I mean, you can point at technologies that are probably five to 10 years out, and you just say, that is just a wild card. I mean, if everybody got into VR, and what we were doing right now with this podcast was a VR type of obsession. And you're seeing stuff like that in the pandemic. You're seeing virtual concerts that are attracting a lot of concurrent viewers. And so, whether you're hanging out watching a virtual rendition of Old Town Road, or you're doing a podcast, and then you do it live with people attending that way, there's a potential for that to increase significant demand. And as that continues to grow, who knows? That's a question Mark that you could say, well, it could peter out as a trend. People might actually start valuing in-person interactions, even more post pandemic. Or we could see ourselves being more and more virtualized because, "Hey, look, my house is comfortable. I'd rather not put shoes on."

Yevgeniy Sverdlik, DCK:

Microsoft just announced a big new project in Atlanta. From being on Facebook site selection strategy, from in your current role also, you're always looking at different markets, examining them for their potential. How do you see the Atlanta market? Also, how does an entry of a large hyperscaler like Microsoft affect the wholesale provider market in the Metro? Once a big hyperscaler has a new project in the market, does that provide impetus for other developers to enter, to maybe provide services around that?

Tim Hughes, Stack:

I mean, I'll just start by saying, we're in the Atlanta market. We have a facility that actually it's... All of our facilities are great. This one's actually particularly interesting, and a great facility. And we had expansion capabilities in a couple of different positions in the market.

Tim Hughes, Stack:

So, we certainly believe in Atlanta. And I think the reason that you believe in Atlanta is because, you look at the East Coast and you ask yourself, "Well, where else is there, besides Northern Virginia?" And the New York, New Jersey market has been plateaued for a little bit now in terms of additional development. And I think that has to do with a number of different factors in that market. And so, people have been pointing to Atlanta as the Southeast hub from a data center perspective. And we think that that's an accurate thesis.

Yevgeniy Sverdlik, DCK:

And partly because of all the connectivity there, right? There's some major carrier hotels. There lots of fiber in the ground.

Tim Hughes, Stack:

Definitely. I mean, when we were deploying our edge network at Facebook, it was one of our top six markets in the United States. So, you had Northern Virginia, you had Northern California, you had Miami, you had New York, you had Dallas, and Atlanta was probably right in there with LA and Seattle, and Chicago. And, it's certainly one of the most connected cities in the country. And it's important because it's really close to Florida, which has a lot of natural disaster risk. And Florida is a significant hub, not just for the Southeast, but also for South America and Central America. And so, there's an important position for Atlanta from that perspective.

Tim Hughes, Stack:

Additionally, and very importantly, it's a large metro in and of itself with a lot of people and a lot of large companies. And so, typically if folks are leasing space for their data centers, they want to be close to their headquarters unless there's a redundancy reason or a significant financial reasons why they would want to do otherwise. So, Atlanta is definitely appealing from that perspective as well.

Tim Hughes, Stack:

The question about the impact of a large scale, a hyperscale, or self-performing in a market, it's really interesting because you have a wide range of outcomes depending on the market. I mean, you can look at a couple of markets in Iowa, for example, that have had significant developments from a hyperscaler perspective. And you're not seeing any movement from the rest of the industry in there. You're seeing hyperscalers clustering there for sure. But you're not seeing the wholesalers come in afterwards.

Tim Hughes, Stack:

Now, on the flip side, a place like New Albany, which until Facebook did their site selection effort there and eventually chose it as one of their us facilities locations. And that was I think, seven years ago. Since then, it's grown significantly from a hyperscale perspective. I think essentially all of them are there at this point. And you're also seeing movement from a wholesale perspective. We have a position there. There's a couple others in the market that are making moves. And I think that's more of a dynamic between New Albany is centrally located in the US population zone. Whereas Iowa, is less centrally located from US population PR positioning perspective. And so, there's definitely dynamics that you see in these different markets that have a lot to do with things outside of just the hyperscalers entering.

Yevgeniy Sverdlik, DCK:

So, a hyperscaler on its own, maybe not enough, but hyperscaler plus other factors in the market, maybe enough of a reason for a wholesaler to enter.

Tim Hughes, Stack:

Absolutely.

Yevgeniy Sverdlik, DCK:

I'm curious also about, so Microsoft has the last few years have been building out their data center, their new cloud regions with availability zones, which they didn't use to do that. They just started to do that recently. And so, when they enter a new market, even if they're entering on their own, I feel like that should also provide opportunity for a company like Stack or anyone else that you guys are competing with to pitch to Microsoft. Do they use a mix of build and lease in a market for the express purpose of being able to enter a market immediately with multiple availability zones? So, you build your core campus on your own, and then you lease out the capacity, or you lease capacity to have multiple locations.

Tim Hughes, Stack:

Again, not able to speak for any client in particular. But what we're seeing as a trend in all of the cloud players is this movement towards these regional positionings with multiple availability zones in the region. And they all have different names for it. But essentially, the architecture is the same in which you have, say Northern Virginia as a region, and then you have a number of locations within Northern Virginia that you can deploy that follow pretty strict guidelines on physical separation and network diversity to allow for some pretty significant levels of redundancy in a given region.

Tim Hughes, Stack:

So yeah, we're seeing that in a number of markets for all of the cloud players, having interest in creating the right level of that availability zone diversity. And there are certainly markets in which they take an approach in which they will do self-perform on some of it, and then they will lease capacity on some of it. That's certainly a dynamic that exists. Some markets, they just decide to lease at all. Some markets they decide to build everything themselves.

Tim Hughes, Stack:

And so really, I think what we want to do, a product that we want to create is, that those cloud clients of ours can really just look at our facilities as an extension of theirs. We build a high enough quality to their standards in an efficient enough manner that they can look at us and say, "Okay, do we want to self perform, 10 miles down the road? Or do we want to lease at a property that a provider like Stack already has? And so, that's certainly something that we look at in markets and we say, "Okay, where are opportunities to collaborate with those cloud providers and ensure that we're working together to develop additional availability zones for them."

Tim Hughes, Stack:

I mean, what you don't want to do is is create problems for your clients by trying to... While we want to predict what our clients want to do, we want to have good enough relationships with them to say, "Okay, you want to develop here? Cool, we're going to develop here. Cool." And actually coordinate really closely as much as possible, as much as they can divulge to us our development strategies in tandem with them.

Yevgeniy Sverdlik, DCK:

When you guys do decide which market to go into next, do you usually already know that one of your clients is going to want to be there, or is there some element of risk in that decision?

Tim Hughes, Stack:

There's normally an element of risk in the decision. I mean, these are companies that are in a very competitive industry that we have great relationships with, but we recognize that they're not going to be able to tell us everything. And that's okay. I mean, I've been on that side of the table. And I always tried to bias towards as much transparency as possible and just say, "Hey, look, I can't tell you any more than that." But we just try to ask the questions and say, "Hey, these are markets that we're looking at that we think are very interesting, that we see potential for growth. Can you give us any feedback on that?" And oftentimes, they'll give us as much as they can. And then beyond that, we have to certainly take on additional risk.

Tim Hughes, Stack:

There's very rarely a case in which somebody is saying, "Hey, go there and we'll lease that space out for you." That's just not really where the market is. They have a lot of options in most of these markets. And so, we have to be on top of our game and create a product both from what it is we're building, but where we're positioning in every single market at any given time our efforts and our capital, so that we can hopefully have an opportunity to work with them in any market that we're entering in.

Yevgeniy Sverdlik, DCK:

They try to keep that information pretty close to the chest. And you mentioned this for competitive reasons. What does one hyperscaler gain by keeping their plans to enter a certain market with the data center campus competitively?

Tim Hughes, Stack:

Well, I think there's two different things going on. One is you work at a company like that, you have to be very sensitive to everything that you say at any given time. I mean, there are so many different stakeholders involved in any of these decisions, especially local stakeholders that you're working with really closely, oftentimes to put these dreams into reality from a new data center campus perspective.

Tim Hughes, Stack:

And so, you never want to, until you know, you don't want to commit to something because there's a lot of decision-making that happens with these. And you want to make sure that you are really honest and straight forward about what it is that you're going to be able to accomplish and what you can commit to at any given time. And then when you can commit, you can go in. And so, just booking it from that side of the equation.

Tim Hughes, Stack:

These are very large companies that are in the spotlight, hundreds of thousands of articles in a month, I'm sure get printed about all of these companies. And so, they need to be thoughtful about the decisions that they're making in the commitments that they're making at any given time, just in general.

Tim Hughes, Stack:

And then in terms of, working with us and competing in the cloud space, for example, I think there's... Let's take us out of the equation and just look at what they're trying to do. They have a lot of clients in these markets, oftentimes that are saying, "Hey, can you deploy additional capacity, or can you deploy new availability zone? Or can you come to this city or this market?" And so, they're working with their own clients to meet their needs. And so again, they can't over commit to any of them. And so, even from a secrecy and competitive nature perspective, they need to make sure that they're able to commit to their clients what they can actually accomplish.

Tim Hughes, Stack:

And then finally, from a competitive perspective, I think that there's what we've seen over the last couple of years, especially is, as the cloud landscape has become increasingly competitive, you see a differentiation in service to proximity. And so, the closer that folks can get to their clients and deploy to their clients, that's a differentiator for them as their cloud services. And so, that's kind of some of their secret sauce. And so, they want to keep that I think as closely guarded in terms of where they're going until they actually have a firm commitment and make an announcement. And then for us look, I mean, we're there to provide services. And that's an important piece of it, but they have a lot of other factors to it.

Yevgeniy Sverdlik, DCK:

So, there's a market with lots of potential cloud users. If I am Amazon and I can build a data center there quickly or ahead of others, I have an advantage in getting those users to use my services, just because the performance is going to be better for them.

Tim Hughes, Stack:

Yeah. Or folks, just want to be closer to the infrastructure that they're deploying. Or perhaps there's local guidelines or regulations that they have to be in compliance with that requires that proximity.

Yevgeniy Sverdlik, DCK:

And speaking of proximity, you were involved in Facebook's edge computing strategy or edge network. Can you tell us a little bit how they go about building on their edge network?

Tim Hughes, Stack:

I can paint a pretty broad picture of what them or any of these folks are trying to achieve, which is the interesting dynamic that I think is hard to pick up on, is that proximity is important certainly. But physical proximity to a user is only as useful as the interconnection point. The proximity of the interconnection point to that user and to the data center that user is trying to connect to.

Yevgeniy Sverdlik, DCK:

Interconnection point between a last mile network and your infrastructure.

Tim Hughes, Stack:

I'll give you an example. I mentioned Miami earlier as a very important Latin American hub. That's shifted. But 10 years ago, I think the majority of LATAM traffic, South American traffic certainly was exchanged in Miami. And that's changed as facilities and exchanges have been built in South America. And that region continues to flourish from an internet perspective. But at the time you could build a data center that served the application in Argentina, for example, but the Argentinian carriers oftentimes would be exchanging traffic with you in Miami. And so, building a core data center closer to the physical user, actually would punish that user because you would have to connect in Miami, and then send the traffic back down to South America to perform whatever that data center was supposed to be performing for that user.

Tim Hughes, Stack:

So, that's an interesting dynamic where you have to be close to where traffic is exchanged. You don't have to be close to where the human is, that is creating that traffic exchange in the first place. Again, you see this countless times in all sorts of places.

Tim Hughes, Stack:

I think Europe is another example. The FLAP, Frankfurt, London, Amsterdam, Paris, which gets a lot of press these days. 10 years ago, it wasn't as much. But you would have questions around, okay, well, how do we better serve a place like Poland? And it's well, actually, it's to deploy more capacity in Frankfurt. Counter-intuitively going to Warsaw, wouldn't actually help most Polish users. That is again, is how that dynamic has changed. But the time the majority of Polish traffic was being exchanged in Frankfurt. And so, building more capacity closer to Frankfurt would be better for that end user than building capacity in Poland.

Yevgeniy Sverdlik, DCK:

Because you can reach more networks of various providers in Frankfurt than you can in Warsaw. Right?

Tim Hughes, Stack:

Yeah. And there's a really a physical component to it as well. Let's say you're a Polish carrier, and you have for the last 20 years been exchanging traffic with the majority of the content networks in Frankfurt. And so, you've built up all of your infrastructure in Frankfurt. And you've built up all of your fiber back haul from your various cell towers, going to your aggregation points to Frankfurt. That's where you exchange traffic.

Tim Hughes, Stack:

And then let's say one of those content providers comes and says, "Hey, we want to actually exchange traffic with you in Warsaw." Well, as a carrier, now you have to build out capacity to that Warsaw location. You have to take all of your aggregation points and back haul it through fiber to that. And maybe you only had a one gig connection. That's not a connection that really people use anywhere, but let's just say you only had a one gig connection in Warsaw, and you had a hundred gigs of capacity in Frankfurt. Now, you're having to build a hundred gigs of capacity and Frankfurt. And so, from a capital deployment perspective, a lot of times for carriers, you want to lean on your legacy infrastructure on where you've been deploying it. And so, it's less about where does a content provider want to go, or a cloud provider want to go, and more about where can a cloud/content provider work in tandem with an internet service provider to find the best place to exchange traffic and serve that traffic from a data center.

Yevgeniy Sverdlik, DCK:

What does that mean for decisions by someone like Facebook about where to put their edge computing nodes versus their core service?

Tim Hughes, Stack:

Yeah. You primarily look for where you can exchange traffic most efficiently. And that's looking at three things. Typically, one is, where is the internet exchange? Because the internet exchanges around the world have a ton of traffic flowing through them. And they're extremely valuable because they're a centralized point where you can plug in whether you're a content provider, a cloud provider, an ISP, or just somebody that's kicking it from the nineties version of the internet and you can exchange traffic on the internet. So, you figure out where those IIXs are, and then you say, "Okay, cool. Beyond the IIX, where are the primary peers that you want to connect to? Where are those large ISPs? What facilities are they in? Are they in multiple facilities? How much capacity do they have in those facilities?" And then you also looking for where you can interconnect with any of the other folks that need to connect through paid carriers.

Tim Hughes, Stack:

And so, you can have you're not just exchanging traffic with your Polish ISP, but you're also exchanging traffic with ISPs behind that ISP, right? Because not everybody, especially the smaller folks can make it all the way to Frankfurt. So, oftentimes, they'll pull their resources and aggregate their traffic. And you can connect to them through, what's typically called a transit provider, which is a third party intermediary between you and other content providers or ISP.

Yevgeniy Sverdlik, DCK:

So, optimizing the peering infrastructure, the traffic that is in your infrastructure. All the hype that's been happening around edge computing in the last couple of years, this is different kind of edge.

Tim Hughes, Stack:

So, the edge is actually lowercase E edge. It's simplistically defined as the point in the network in which you're interconnecting with somebody else. And so, a content provider or a cloud provider has an edge with an ISP, that's typically in one of these exchange points, right? Let's say in Frankfurt. Now, that ISP also has an edge with their client. So, you have where they're exchanging traffic as one side of their edge, right? Where they're exchanging traffic with a cloud provider or a content provider. The other side of their network has another edge where they're actually connecting with their clients. They're paying customers. That is the edge that's currently being talked about, where people are trying to optimize very hyperlocalized services for that. And that's a very difficult problem because it's so dis-aggregated that you have to be very specific in the applications that you're trying to optimize at that point. Otherwise, you don't have the right scale to actually properly optimize it and you can actually create problems for your clients.

Yevgeniy Sverdlik, DCK:

I want to go back to the hyperscale build versus lease question. So, why don't hyperscalers, they obviously have the best teams, the best designers, the best operations teams, they are very smart people working on their data centers. They build these state-of-the-art facilities, super efficient. Why don't they just do it all on their own? Why don't they just build their own data centers everywhere instead of leasing?

Tim Hughes, Stack:

Yeah. I think there's two primary reasons that they're leasing right now. One is relatively straightforward and intuitive to folks in the industry, which is there are some places that you want to deploy capacity as a hyperscaler, where you don't have the scale from the demand side, from your application or from the cloud that you're building there, that you actually want to self perform and build your own facility. And so, oftentimes these are in international markets, or these are in traffic exchange points, where it just doesn't make sense to self-performance. So, you're going to lease in those places because you're not passing the threshold to what we talked about earlier, in which you're able to actually capture the economies of scale of deploying your own facility.

Tim Hughes, Stack:

Additionally, if you start trying to build and deploy in every single one of those locations, you are stretching your team very thin, from an operations perspective, from a design perspective, from an engineering perspective. And so, you want to work with partners in those locations for those instances. And that's a pretty significant portion of the demand, especially as the proximity wars, that competitive aspect of the cloud industry right now in which folks are trying to get closer and closer to their clients as that continues to mature. So, that's one piece of the puzzle.

Tim Hughes, Stack:

The other is, oftentimes the leased facilities are really a release valve I guess, pun intended for capacity requirements that weren't necessarily predicted. The development and deployment cycles for data centers are very long. Typically, you're seeing anything from, starting the project to coming online, 18, 24 months. People say 12 months. They're kind of shaving some stuff off on the front end, on the back end of that, to make that 12 months look good on a schedule, but it's typically 18 to 24 months.

Tim Hughes, Stack:

And so, physical building development, the procurement of energy, mobilizing large construction crews to get this done, working with local stakeholders to ensure that you have everything that you need from a permitting perspective, that's a lot of work and a lot of time. And so, if you are predicting your demand 18 to 24 months on a rolling cycle out into the future, sometimes you're going to miss low. And oftentimes, that's where we're filling that gap, where folks are saying, "Hey, you know what? We have this demand cycle that is on this weekly increment or even daily increment." Because especially from a cloud perspective, you can deploy software pretty much instantaneously on the cloud.

Tim Hughes, Stack:

And so, if you and I came up with this killer podcast app that a billion people decided to use tomorrow, and we deployed it on one of the large cloud providers, they would need to spin up additional capacity for that. And it's possible that they would need actually additional physical capacity to do so. If that's the case, then they might not have that 18 to 24 months of luxury to deploy additional data center capacity, which means that Stack Infrastructure can step in and say, "Hey, we have 200 acres in Manassas. And we have plenty of capacity available there. And we have a bunch of electrical interconnection capacity as well. And by the way, we've already graded the pad. We can be up in six to nine months." And then folks are interested. And so, you're oftentimes as a partner filling the gap for the demand predictions that sometimes fall short in the industry.

Yevgeniy Sverdlik, DCK:

So, it helps kind of soften the unpredictability, the unpredictable nature of their demand. So, they take a smaller risk by going into a location.

Tim Hughes, Stack:

Yeah. I think it's less about them taking a smaller risk and more about them saying, "Hey, look, here are our various model predictions for demand over the next 24 months, 60 months." Whatever window they're looking into. "And here's the 10 percentile view of that. And here's the 90 percentile view of that." And they cut off the ends just because you start to see some really wonky stuff. But they have some very smart data scientists putting together models and they say, "Hey, here's the 10, here's the 90. From a capital allocation perspective, perhaps we should develop to the 60, which gives us maybe a little bit of buffer."

Tim Hughes, Stack:

But the difference between the 60 and the 90 can be a pretty significant amount of capacity. And so, what they'll do, is intelligently continuously refresh that model and feedback with their teams cross-functionally to that point earlier that are actually procuring and deploying these data centers.

Tim Hughes, Stack:

And if, for some reason that does spike to a 90, and it does so in a very small period of time, they don't have the ability to respond to that as rapidly as they probably want to, and self-perform to that. And so, it's all about modeling. And it's all about trying to predict the future as accurately as possible. And you're never going to get it a hundred percent right. And so, the question is, are you over or are you under? Is it within your buffer? Is it outside of your buffer? And how do you address the gap?

Yevgeniy Sverdlik, DCK:

Okay, Tim, that's all I have. Thank you so much for your time, man.

Tim Hughes, Stack:

Absolutely. Thanks. This has been a lot of fun.

Hide comments

Comments

  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
Publish