Soluna & Metrobloks: Powering AI at Project Kati 2

Welcome back to Clean Integration, a Saluna podcast. I'm John Belizeir, CEO of Saluna and your host. Today we're talking about something important for Saluna and for the broader AI infrastructure market. We recently announced they joined venture with Metroblocks to co-develop Project Coddy 2, a 100 megawatt plus AI in high performance computing data center, our Win Powered Project in Southeast Texas. This partnership combines secured, renewable power with AI-ready design. To talk about why this matters, I'm joined by Ernest Purpescu, CEO of Metroblocks, Ernest and his team have enabled more than 12 gigawatts of global data center capacity, and they're building specifically for the next generation of AI workloads. Ernest, welcome to the show. Thanks for having me, John. Glad to be here. Well, I want to start by zooming on the joint venture, and before we dive into that joint venture, can you give the listeners a quick overview of your background and what led you to start and build Metroblocks? Yeah, of course. I started in the data center development industry, I want to say about 13 years ago, maybe 14 at this point. I started with AWS planning global data center capacity, and then moved over to managing the teams that I was triggering, especially responding to the demand signal by leasing capacity, identifying greenfield, development opportunities, entitling, permitting, powering, all the way through to the construction and the design of the data center. Did that for AWS for about three or four years enabled upwards of 10 gigawatts of capacity in the Americas, Canada, the US, and Latin America? After I took on a similar role with Meta Facebook at the time, it was a global role, which was appealing and predominantly focused on Europe and the US. I knew I wanted to try my hand at the entrepreneurial route, but before doing so, I wanted to experience the industry from a different perspective. I had been leasing, buying, developing capacity for the hyper-scalers. I wanted to work as a developer, building and selling capacity in large part back to my former employees. I joined Iron Mountain. It was interesting, informative for a couple of years. About a year and a half ago, I founded Metroblox with the intent and idea to bring next generation AI-ready capacity focused on urban and filled opportunities, but also opportunistically exploring larger scale development projects like the one we're involved in. How's it been so far running your own business versus being in a much larger enterprise? Yeah, it's been everything I hoped it would be and more that's certainly been the most challenging and rewarding professional experience of my career. It's been equally tough frustrating at times, and ultimately it's a very personal decision. I think the timing was right for me, and I continued to be excited at the positive momentum we're building, and I continued to be excited at the relationship I'm building. Again, the one that you and our companies are in the process of Chris Lab. Give me your perspective. This news, at least for Soluna, is an important piece of news around this partnership. Soluna Metroblox recently announced that Ajay Venture to co-develop project Kite2. It's going to be a big campus. From your perspective, why does this partnership make sense right now with everything that's happening? Yeah, I think it's a very complimentary partnership. Soluna's track record of contracting renewable energy is both appealing from an end user occupied perspective, and also from an ESG perspective, the right thing to do. Being a good corporate steward is becoming more and more important. There's a lot of negative sentiment around data-centered development. Doing it the right way is not only the right thing to do, but also the smart thing to do. The grid is stretched. I think tapping into existing renewable projects is smart, and sort of having a behind-the-scenes solution is efficient. I can't think of enough benefits, which are certainly what will make the relationship appealing, and I think the location, the size, the timing, of the power delivery. I think the partnership in terms of Soluna's expertise and hours is complimentary, and hopefully this moonshot, and this opportunity will certainly from the feedback we've gotten from customers, certainly appears that we're doing the right thing. Yeah. I mean, in this deal, Soluna brings control and our entitlements, a site-controlled power entitlements, or kind of good access. When energy, we're focused on electrical equipment, and the energy development expertise, of course, are as that listed platform for capital formation. Can you walk us through what Metroblocks brings to the table for the benefit of the listeners and some of our shareholders that might be online? Yeah, I think Metroblocks' team, you know, sort of brings a technical state-of-the-art design industry relationships, both in terms of data-centered development, construction, long-leaf equipment procurement, and also customer relationships. I think, you know, so having those two ends meet is complimentary, and we'll hopefully prove to be very successful. I want to switch to talking about the market now. Everybody's talking about AI and the data center industry being one of the sexy industries in a while, and energy being sexy for a while. What's changed in the last few years that made AI infrastructure fundamentally different from, let's say, traditional cloud or just data centers in general? What's different this time? Yeah. I mean, a few things are different. I think the collective sort of humanity's bet on the value of AI being ubiquitous and life-filtering, you know, certainly makes it a hot topic, but I think mechanistically what's changed from a data-centered development perspective is, well, there's a few things that the first is the size of these AI and well deployments are constantly increasing in large part in response to densifying hardware, racks, servers. And in second part due to the large sort of computational requirements to operate these large language models. So it's been really fascinating time in the industry, and in an industry that was already experiencing, you know, 30, 40, 50 percent year-over-year growth, AI has just exponentially increased that. The other thing that sort of makes AI, ML data-centered development interesting is that it's in large part location and nasty and doesn't necessarily need to abide by the same rigor that, you know, more traditional compute and storage and sort of availability, zone, public cloud deployments needed to adhere to. And I think what that's done is it's opened up the market to where instead of looking at sites, you're now chasing power and developing data centers where power is available either from the grid or off-grid through islanded onsite generation solutions or behind the mere solution. So it's really been game-changing in terms of opening up the country and the world to, you know, large-scale AI ML points. Yeah, I mean, I always find it fascinating how acute pain creates innovation, right? Big sort of shifts in the way people think about infrastructure. You know, the internet did that for telecom and now we're seeing it for physical or digital infrastructure and power. You know, you're out there talking to customers for your projects and for the projects we're working on together. You know, what do you see as the real bottleneck in the market today, how compressor expectations compared to, let's say, three years ago and if the customer's looking for 50 to 100 megawatts in the next 12 to 24 months, you know, how many realistic options exist these days given all the momentum that's underway? Yeah, I mean, to your point, you know, necessity is certainly the mother of all invention and I think the large power consumption has triggered innovation and the rethinking of where and how data centers are cited and how their power. In terms of customers, you know, I think what's also changed is to, I think you made this point earlier, you know, I think you referred to it as a sexy, you know, sort of real estate asset class. I think it's attracted, you know, sort of a lot of interest, some more legitimate than other and, you know, I think there's a lot of noise and that noise is amplified both in terms of securing power from the utilities, but also in terms of presenting. implementing actual actionable opportunities to customers. And I think that the, this market is really interesting because your occupiers, your high credit investment grade, where the occupiers, the hyperscale, you ask Microsoft, Google, Facebook, or a code to a lesser extent, and some of the well-funded neo clouds with sort of investment grade credit support are the ones you're actually competing against as a developer. So they're both your customers and your competitors. So it makes for a leasing dynamic. I think what's happened is there's been an influx of, wanna quote, opportunities, and I think it's sort of mucked up and clogged up the pipeline for the real development opportunities to sort of rise to the top. But I think a partnership like ours is really interesting and sort of exceptional because it combines both renewable, real power work that you've already done, quick time to market with hopefully a credible technical team of industry experts with relationships and connections to both the development community but also the end user occupier community. So I think those types of partnerships are what will hopefully differentiate our project from the many others that are attempting to sort of get airspace that are out there. - Yeah, that's a good segue. We're just talking more about Project Coddy specifically, I mean, you evaluated sites all over the US and looks like you continue to do some blocking and tackling on those sites around the country. How are yours to find 100 megawatts with scalable capacity with power already secured? Tell us what makes South Texas strategically differentiated based on your due diligence and analysis on Coddy and what gave you the confidence that this site could scale beyond the first phase? - Yeah, it's interesting. There are quite a few things and maybe none more important than the relationship that you and I have established and sort of your vision and confidence and your ability to deliver power is probably the first and biggest differentiator. Second, what's really interesting is there's this concept of shared faith zones and data centers are all the tends to purposes of utility, a vehicle to conduit for, in this case, sort of large language model training. There's a concentration of capacity in West Texas, central Texas. I think what makes East Texas interesting is the fact that East Texas and the state of Texas being so large doesn't share that sort of shared faith from a disaster recovery perspective is something where to happen. So that ended up itself makes this location interesting. I think the other thing that makes it interesting is proximity to the Mexican border and its connectivity tissue to between the US and Mexico which I think also makes it really appealing. And then the other thing that I think from a legitimacy perspective, I think the partnership with an experienced renewable developer and with access to renewable power, land, and the speed to markets, or what makes this asset pretty unique. Got it. I want to talk about AI ready now. I think it's important when you're building data centers to have a eye to the eventual application. And one of the things we're doing together on this site from a technical perspective is really designing a facility that's flexible and available for AI and we're calling it AI ready. What does that mean in practical terms for the listeners who may be novices to the space, perhaps talk about densities, how does it compare such additional hyper scale data centers, flexibility and chip architecture, the things that you would perhaps talk to a customer about. - Yeah, what's really interesting is the development cycle times for a physical infrastructure and sort of the chip hardware are so incompatible. And the development cycle time is 36 to 48 months, right? And in our case, it happens to be roughly half of that because of some of the pre-work that you've already done. But typically, it's 36 to 48 months. The hardware lifecycle is, if you look at Nvidia, they're rolling out chips every 12 to 18 months. And if you overlay Nvidia with AMD and Micron and some of the other chip manufacturers, you can argue that the lifecycle for new hardware and more dense, more efficient hardware is some 12 months because they sort of leapfrog each other and are continuously reinventing what's possible. So in my mind, what AI ready means is having the flexibility to accommodate an evolving, densifying rack type. It means building flexibility into your design where you can cool, which is essentially what data center for structure in large part does. Right. Cool racks ranging from density per rack position, sort of traditional compute storage, which was 30 to 40 KW per rack position to the latest iteration of chips, which is 150 KW per rack position to the newest generation of chips, the GB 300s, and thereafter, which are claiming to require over 250 KW per rack position and pushing a megawatt per rack position, which is insane. That's sort of more of the law, the sort of exponential growth. We certainly don't seem to have peaked in terms of rack densities. So I think that's what AI ready in my mind means it means building flexible, adaptable infrastructure to accommodate this very fast evolving landscape of hardware densities. And what do you see developers, perhaps mistakes, other developers are making when they try to retrofit to these traditional data centers to AI? Yeah, I mean, it's interesting because retrofitting and I've been doing this a long time, and the sort of Moby Dick, the white whale, has always been the concept of retrofitting. It sounds great in theory. It is next impossible to execute practice for a few reasons. And it's very difficult. So the mistakes I'm seeing developers making are not being forward looking, not building the flexibility in their design to accommodate an evolving landscape. And in some cases, they can't, right? In our case, we've got plenty of land and plenty of power. So we can afford to make design decisions to allow us for margin of error and to accommodate, again, like I said, a very fast evolving hardware landscape. In some cases, you can't, you don't have the land, you've got the power, you've got to go to story, we're trying to cram as much capacity as you can into the existing footprint. And I think you end up making some short-sighted decisions which expose you to alternate uses in terminal value risk. Got it. No, very exciting. And for the listeners, KW is killer Watts. For those of you who haven't heard that, turn before. I want to talk about capital and big picture. What makes this JV structure attractive to potential infrastructure investors? What's unique about it, perhaps, from your experience in engaging with these types of investors? And what will they see attractive about investing in this project? Yeah, I mean, it's really interesting because like most realists say, what makes the infrastructure attractive is the tent. You know, sadly, I think we're the conduit of vehicle to enabling, which sounds insane to say, because you're talking about billion-dollar development projects, but when you compare it to the chip hardware investment, which is usually in the realm of 4 to 5x, the infrastructure investment, and refresh every three to five years, it pales in comparison. So my rough order magnitude is that for every billion dollars invested in infrastructure, there's somewhere between five to 10 billion dollars invested in hardware, refreshed every four to five years, which over the lifecycle of a data center, it's eight refreshes. So you could be investing anywhere between 40 to $80 billion in hardware, and eight particular data center over a 40-year lifespan. So, and then that is also the hardware, which then in turn monetizes into what should be revenue generating. So what makes this attractive is, I think, the combination of the partnership, the renewable development expertise, the data center development expertise, customer relationships. But ultimately, having settled that, I think that legitimizes the investment. I think what-- and then that attracts tenants that are high quality credit worthy investment-grade tenants. And I think what makes the investors most interested is the combination of all of those factors, which makes for an exciting development opportunity. and one that has attracted a lot of customer interest, and I think that customer interest will translate into investor interest. - Yeah, for sure. - I have one closing question. If you're an AI operator right now, a potential customer listening to this pike has, why should Goddy-T be on your radar? - You know, one word comes to mind, legitimate. You know, in the world where there's a lot of noise, a lot of pretenders, a lot of imposter's, who are out there selling powered land, which sadly is sort of built on a deck of cards. I think this project is legitimate. This project is coming together of legitimate organizations with large aspirations. And I think the legitimacy and the speed to market are why you should pay attention to this project. - We're in this. Thanks for taking the time to join us today on today's episode and for building this alongside us at Project Coddy-2. - Exciting to be a part of this, and you'll appreciate the opportunity to share my thoughts and insights and opinions in the industry and on our relationship. - Well, folks, Project Coddy-2 represents something bigger than this single site. It's a signal that AI infrastructure is increasingly defined by energy access, speed, and execution. By combining secured renewable power with AI-ready design, we're creating a platform built for what's the market that it's actually looking for. And if you're an AI operator and an infrastructure investor and want to learn more about Project Coddy-2, check out our show notes for more information and ways to get in touch. Thanks for listening to Clean Integration. We'll see you next time.