Marc Ganzi
Analyst · Cowen & Company
No, you always make sense. And I think you're skating to where the puck is going, not to where the puck is. I want to sort of break your question down into 2 pieces, Michael. One is just to talk about the direction of travel on power. And then the second, I do want to talk about connectivity because you've nailed it, right? Connectivity is really critical. The capillaries that connect, obviously, interconnection to data centers, and then how this correlates to AI and where are those big language-based models being built and ultimately when we move from you know from training into inference, those locations become more latency sensitive, so we can explore that for a second.
And I’ll be mindful that there's other people in the queue that have questions. This is a topic you and I could talk a lot about. Remember, data centers have sort of, there's kind of 2 screens to this. The first is, do our customers trust us to build their data centers and to continue to lease capacity from us? And what kind of workloads are they leasing from us? Coming out of this quarter, we had contributions from all 6 of our major data center platforms globally. So we spent a lot of time this last week aggregating that data and understanding what our customers are doing.
And so what's really interesting to me is that there's such segmentation now between cloud and AI in those workloads and ultimately workloads that are latency sensitive and workloads that actually are very location sensitive from an AZ perspective. And so private cloud, public cloud, edge workloads, enterprise workloads, all of those workloads, Michael, are evolving in real time. And data centers are evolving too. And so I think that some of these locations that are less latency sensitive are some of these locations that can be 200, 400, 800 megawatts. And those are those training models in AI. And provided you have good fiber connectivity, those locations can be a little less, shall we say, latency sensitive.
Then there are, as you move in a Generative AI and you get into active workloads and active applications, it starts to follow the same model that the cloud followed, which is, you and I both know we're in the 11th to 12th year of the cloud. And so we're seeing a lot of those locations in those AZs are now really important in places like Goodyear, Arizona, in Atlanta, Georgia, in Columbus, Ohio. And certainly like Reno, Nevada is an alternative to Santa Clara. And so there's this whole next generation of cloud workloads that are showing up big in its scale, but they're not traditionally in Virginia, they're not traditionally in Santa Clara. And you see that the customers are navigating to different places.
And then obviously you see what's happening at DataBank on the edge side. DataBank had a fantastic quarter, one of the best quarters in history, and that company continues to deliver what we call hyper-edge workloads, that half megawatt to 10 megawatt workloads, where the cloud is obviously moving to secondary and tertiary markets. We see AI following a similar footprint, but the challenge against how you build AI is very much correlated to where you can get power. And then of course, there's the self-perform where our customers are going to perform their own work, and ultimately the work that we're going to perform. This is a very complex matrix, because it's a decision tree at the end of the day, Michael, the way I think about it.
And the first decision is, are customers going to trust us to build the workload, are they going to self-perform? Check. Second screen is, is this really for their cloud products, or is this for their AI product? And then of course the engineering standards, and ultimately the GPU standards, and the design standards, and the cooling standards change and deviate a little bit. And then ultimately do people really value sort of having a Tier 5 experience where those workloads need to be highly secure and perhaps even in private cloud, aka what Switch is doing. Switch has had a phenomenal first quarter and the second quarter is even lining up to be better. The good news is, in our world, we don't have to choose. When we own powerful platforms like Vantage, AtlasEdge, DataBank, Scala in Brazil, which had a great quarter, we're seeing all of these workloads manifest itself all across the globe. As you saw in our slide, you see photographs of data centers from all around the world.
I don't see that the customer is the constraining factor, Michael. I see that power is really the constraining factor. And that's going to become more evident to you and to the rest of the investor community over the next 2 years. It's not obviously, from my perspective, Michael, it's not new news. We started talking about this over 2 years ago at the Berlin Infrastructure Conference when I told the investor world, we're running out of power in 5 years. Well, I was wrong about that. We're kind of running out of power in the next 18 to 24 months. So we started 2 years ago working on this power problem. So it's not new information to us. And as I said on our call today, the 2 point plus gigawatts that we're building today, shovels in the ground, all of those commitments with our customers have power.
Power in place will serve letters. Those aren't hope data centers. Those are actually data centers that are committed, being constructed, and customers are moving into it. I do look around the corner and I look at that next 5 plus gigawatts of opportunity and we're going to have to get more creative. And the way we get more creative is; one, we try to locate certain of those big AI data centers and locations that maybe are less latency sensitive. We try to co-locate those opportunities closer to renewable energy and we try to create energy independence or grid independence. And those are the things that we're thinking about. So the next generation of data centers are perhaps going to be in different locations.
Now, how do we create that customer experience? We create those customer experiences with low latency, big, big, big pipes in terms of the dark fiber that we're bringing to those data centers. And I'm not talking about 4 pairs, 12 pairs. We're talking about hundreds of pairs of fiber with redundant routes. And this is where, as you've highlighted, Zayo comes into the mix, and what gives us a lot of confidence to sit with some of our key customers and say, look, we'll deliver you the data center. We can deliver the fiber, and then the next sort of key is can we deliver the power. Now if I can wrap that all up in one bow, we actually have InfraBridge, which as you know is an infrastructure provider. We actually engage in renewable energy already as a firm. And so our opportunity set is fusing the hard work that we did at InfraBridge, some of the hard work that we've done at Zayo, the hard work we've done across all of our hyperscale and private cloud data center operators to bring a holistic solution to customers.
And now that's finally manifesting itself. Now, the real key is at the asset manager level, we got to make that manifest itself in the capital we form, the fees that we generate and the carried interest that comes commiserate with creating these great ideas and bringing it all together. I am very happy with what's going on at Zayo. We've seen a significant uptick in the bookings there, particularly with the hyperscalers and some of the web scale routes. There's great opportunity there, and they're going to need us. And it's not just for Zayo, it's the whole fiber industry in general is going to need more new routes, low latency routes, and of course heavy strand count. And that's the way you bridge the gap in terms of creating low latency environments for AI workloads. So it's a combination of a lot of things.
This is -- this situation Michael, and I'm sorry this is a long-winded answer but you asked a complicated question and we're going to dig into this in our Investor Day. It's more complicated to build a data center today than it was 2, 3 years ago and it's going to get increasingly more complicated. And I've been saying that for the last couple of years, but now finally everyone's paying attention. And it won't get easier. I can share with you that it'll get harder. But I like it when things get harder. When it was harder to build towers 20 years ago, we were up for that challenge. When it was harder to build small cells 15 years ago, we were up for that challenge. When it was harder to build small cells, you know, 15 years ago we were up for that challenge. So we got a management team that understands how to work through challenges and the key was identifying those challenges over 2 years ago, which we did. So stay tuned and I think we've got a great story on how we solve problems for our customers.