All right. Who is ready to step up the voltage for this one? No? Biotech guys didn't get that. Yeah. Just had a double espresso and I am running at at least a gigawatt right now, so I think it is a good time to begin. Good afternoon, everyone. I am Mike Colonnese. I head up H.C. Wainwright's crypto and digital infrastructure research franchise, and I am extremely excited and grateful to be joined by such an esteemed group of panelists to discuss one of the most prominent investment themes of our time, AI infrastructure.
Simply put, the companies on this stage have sourced and secured the scarcest asset in the AI value chain today, access to scalable power. Some interesting data points to share with you to set the stage before we begin today. First, as many of you know, the hyperscalers are expected to spend over $1 trillion on AI next year alone. Second, Bitcoin miners have secured over $160 billion of AI colocation deals to date, and the companies on this stage alone have line of sight to over 14 GW of total power capacity. With that context set, I would now like to pass it over to the panelists for brief introductions. Matt, why don't we start with you and work our way down?
Thanks, Mike. Good afternoon, everybody. Thanks for being here. My name is Matt Schultz. I am the Chairman and CEO at CleanSpark.
Fred Thiel, Chairman and CEO at MARA Holdings.
John Belizaire, CEO of Soluna Holdings.
Josh Kilgore, Executive Chairman for Big Digital Energy .
Michael Potter, I am the CFO of Bitdeer.
Russell Cann, I am the Co-Founder and Chief Development Officer at Core Scientific.
Thank you for that, gentlemen. I have had the great fortune to get to know a lot of the gentlemen on the stage today as one of the first publishing analysts on both crypto and digital infrastructure. What I have concluded over the years is that Russell is the best dancer on this stage. All right. Anyway.
What? Wait a second. You have not seen me yet.
Oh, sorry, John. Maybe after tonight, we will change that. All right. Well, Fred, listen. It would be great if you can kick us off to discuss Bitcoin miners' evolution and transformation into AI data center developers, and really the broader market opportunity for the sector here.
Sure. As you think about Bitcoin mining, the focus was acquire power you could turn on as low cost as possible, bring in compute to run to solve Bitcoin algorithm, and then generate as much Bitcoin as you could at as low as cost as possible. So we were constantly being driven by how do you get the latest infrastructure at the lowest cost, optimize your power, and make sure that you are not spending too much to build out your infrastructure.
If we look at Bitcoin mining economics, typically, you would be spending somewhere around $1 million-$1.5 million a megawatt for infrastructure and compute. Very different than the business that most of us are transitioning to or find ourselves in today around AI HPC. One thing that Bitcoin miners have a lot of is land and power that is turned on. Those of you who are familiar with the AI HPC business know that the shortage of operating compute is driven by a lack of available power powering a building that is suitable for running AI HPC.
Bitcoin miners are a very quick path to solving that problem for hyperscalers, frontier model providers, and neoclouds. As I am sure all of my colleagues here can attest, they have been in conversations with all of them, and we are a very quick way of solving a problem, which is getting a lot of AI compute online fast. If you look at a revenue per megawatt basis or an asset value per megawatt basis, it provides this considerable step-up for Bitcoin miners to convert their assets to AI HPC. With Bitcoin price having gone through the more recent evolution since Q3 of last year, we have all had an incentive to accelerate our plans into this transition. That is why we find ourselves where we are today.
Great overview, Fred. Thank you for that. Russell, Core Scientific was the first Bitcoin miner to secure the first large-scale AI colocation deal, I think over two years ago now. If you could walk us through that first contract and how you have seen the market evolve since then, that would be great.
Yeah. Our first deal actually was in February 2024. It was a small air-cooled site. It was a site that we leased, and then we converted it. It was a 5 kW to 7 kW rack. We converted it to 17 kW to 20 kW rack to run the H100 to H200s. That was like a 45-day conversion. The next deal we worked out with CoreWeave, though, we actually signed it in June of 2024, and it seems like a decade and a half ago. If anything moves as fast as Bitcoin, it is actually this AI place because, in that two years and three months we have turned on and we are building like 450, 500 MW of that right now, which is a good place to be in.
But all those other megawatts had to be built from the ground up, and that is basically because direct liquid cooling and air cooling were just not the same thing. The legacy data centers that were air-cooled, it made no sense to try to convert those. We had to build everything from the ground up. You asked about some of the first deals versus right now. I remember very distinctly when we were pricing this, the first deals, it was like in late 2023. We had good hard estimates of around $4.5 million a megawatt to build the direct liquid cooling stuff. By the time we got around to inking the deal in June, that same facility is just under $8 million a megawatt.
Now, the things that we actually turned on this year, the other sites, they were floating around $10 million-$10.5 million a megawatt. Then to give you a little bit of ideas, the stuff that we have turning on in 2027, next year, the stuff that is turning on, is between $12 million and $13 million a megawatt. Now, most of that increase is in labor and then in a lot of the gear that is actually coming in to switch gear, the transformers and stuff. But to give you an idea on labor, if you have a kid graduating high school right now, they need to go become an electrician. Journeyman electricians right now on one of our sites make about $250,000-$350,000 a year, and master electricians are bringing in $750,000 a year.
Labor has gone up, and it is literally one of the key points right now for being a roadblock. It is like, is there power at the site? Is there NIMBY at the site? Then do you have your long lead equipment, transformer, switcher, that kind of stuff, and do you have your labor? Those are kind of the points. Every day, that is what we debate now. The things that were the issues have just gotten bigger. As Fred just said, Bitcoin miners are in a unique position because we have a lot of those things. We have been dealing with a lot of those things. I am not sure if I answered your question, but I tried to give you a little background on the history there of those first sites.
Spot on. Thanks for the history there, Russell. There are two primary business models to monetizing AI infrastructure deployments. We have co-location and AI cloud services. Sam, welcome to the panel. Good to see you.
Hi. Thanks for having me.
Great to have you. This one's for you. At WhiteFiber, you are actually pursuing both strategies, right? You are doing co-location, you are doing some cloud services. If you could talk to us about the differences between the two models and why WhiteFiber has elected to pursue both strategies.
Well, WhiteFiber actually started in the cloud business. We are, I think, one of the first in the sector, in the Bitcoin mining sector, to have that cloud contract, and it was essentially a catalyst for us to spin off the business. We felt, instead of turning Big Digital into an AI infrastructure platform, that it would be better, and there are a couple of reasons for it, to just spin out the business, which is now today called WhiteFiber, which IPO'd last August. There are a couple of technical reasons for that. One of them is financing. We were dealing with a lot of conservative banks who wanted to finance our AI infrastructure business, but were uncomfortable with the fact that there was some crypto in the business.
One of the things we were able to do to provide comfort was just start over, just spin out the business and just make it a pure play AI infrastructure business with absolutely no crypto, so that way these very conservative credit committees would get comfortable on establishing some credit facilities to develop our AI infrastructure. Another reason why we decided to spin out is just to have a more institutional shareholding capital structure.
When we did the IPO for WhiteFiber, we were talking to Bostonian long-only funds. We were talking to very institutional blue-chip shareholders, and these are not the type of shareholders that would be interested if we had crypto. They were interested in AI infrastructure, not crypto. Those are the two main reasons why we decided to IPO the business, and we started with this very large cloud deal.
We felt, to be AI infrastructure, you should really have both the cloud side and the co-location side. We felt that we did not have the technical chops to really create an AI campus. Creating a Bitcoin mining facility is not the same as creating a data center for AI infrastructure. So what we did was we acquired a company called Enovum a couple of years ago. That company, that team, has been doing AI campuses for many, many years in a retrofit format.
They worked for hyperscalers such as Amazon and Microsoft. That team has been doing that for so long, and they just knew how to do that retrofit format. We felt that the skill set to create a data center for AI customers, just so different from a Bitcoin mining facility, that it would make sense to acquire that company. That company became our co-location business, then we have our cloud team, and those are two separate teams and very, very different skill set.
There are some people here on the panel who may disagree with my position that it is just a very different skill set, creating a Bitcoin mine from transforming Bitcoin to AI campuses. But I can tell you that we have done it now. We have taken over mattress factories. We have turned that into data centers. We did that for Cerebras that recently IPO'd. We have taken over a 1 million sq ft facility in North Carolina. We did that for Enscale for an $865 million contract. They are now up and running. They are up and running. We have done that within six months.
That was not because of our abilities as former Bitcoin miners, but because we acquired that particular specialized team that does that thing and have been doing that for many, many years. That is sort of a meandering way of answering the question that this cloud business was the catalyst that started this whole company that is called WhiteFiber today, and it is divided into two. There were technical reasons for it. The first reason was shareholding, the second was financing. With respect to the co-location business, that is very separate from the cloud business. It is a very different skill set. We are very proud with what we have built in just basically one year.
A mattress factory to an AI data center in six months. Pretty impressive.
For Cerebras, yeah.
For Cerebras. Good stuff. Thank you for that, Sam. Michael, you are also scaling both business models over at Bitdeer. Anything to add?
I just think that it is a very different process to do both of them. Because of the size and scale that we are and approaching the business, we decided for Tydal, Norway, it would be better to do a co-location deal. So we signed the deal with Volta, which is providing GPUs for their end customer. It is about 121 critical IT megawatts. And we are building out the data center itself, and the GPU is being handled by our tenant.
That way, we can learn how to scale and build large data centers to the current technical specs that are required, while somebody else is taking care of the GPU and the GPU financing. So we are pleased that that deal went through, and working hard to get the buildings and everything finished for the end of this year and the end of Q1 of next year. At the same time, we feel that, one, we have some smaller sites that do not make as much sense to do co-location alone. It is just not the scale that the big players would like.
And two, to take the land and power that we have worked so hard to build up and lock it all up, all of it, for 20 years, it is almost like becoming a REIT, and it sort of takes the optionality of what we could do with this land and power ourselves. So we are also starting a neocloud or a GPU-as-a-service business, mostly starting in Malaysia right now. Our headquarters in Singapore puts us in the Asian market and lets us contact some leading Asian customers. So starting off with about a 10-MW neocloud site in Malaysia, which we have two customers for.
We are going to be rapidly. We have another 20 MW in Malaysia, and we just announced this morning another 65 MW. We have good customer pipelines there to fill them up. In the U.S., we have a 10-MW facility in Washington, about a 50 MW facility in Tennessee, and about another 40, 50 MW left in Tydal, Norway, that we could do something with as well.
I think we are well-positioned to do neocloud at smaller sites where we can scale and learn the skills and get the software all perfected. We have two large sites in the U.S., with Rockdale being the easiest. We just did a large land purchase. For that, the size that we could build an AI data center may be too big for us to do everything, to build the data center and to do the GPU financing. We would look at doing co-location as a possibility. But again, maybe we do not want to lock up all of the available power in the sites that we have so that we are not locked up our best asset for the next 20 years. That is the business we are working through right now.
For those on the panel who are doing only co-location today, would you consider procuring your own GPUs on future deployments? Why or why not? I will leave that one open for the group.
Yeah. One of the reasons we took majority control of the French company, Exaion, which was a subsidiary of EDF, was specifically because that is their specialty. We did that. That deal closed in March of this year, and they have already started taking contracts, interestingly enough, hosting U.S. inference and training loads in Europe because the delays in getting stuff online in the U.S., there are now people seem very happy to take European capacity, and then we will bring Exaion's expertise in running these mission-critical data centers and building them onshore in the U.S. as we continue to develop the smaller sites we have, which, to your point, a lot of times they do not make sense for a hyperscaler or another neocloud to take them. They are better off just using its inference sites.
All right, so we talked about the history, how the market has evolved, the broader market opportunity, and ways to monetize these deployments. I'd now like to focus on recent deals that each of you have secured, some of the specifics around those contracts, in addition to a question we get a lot from investors, sites within your portfolio that you believe are closest to lease execution and why. We'll go through the whole panel for this one. Matt, why don't we start with you? CleanSpark entered the game, coming out the game strong with a $6.6 billion deal a couple of months ago.
Yeah. Thanks, Mike. CleanSpark has amassed a portfolio of 35 different sites around the country. We're in a number of different states, and we made the decision about a year ago to start to evaluate the highest and best use for the electrons that we currently have energized. What is a massive differentiator is having substations built and live power today because, as we've seen, the barriers to entry and really the bottleneck for development of data centers really is land and power.
So our first site that we leased is in Sandersville, Georgia. We've been operating there for almost four years with 11 exahash of Bitcoin mining. But we built a substation. It's fully energized. It's 250 MW of power. So we met with our tenant and came to understand their basis of design and the reference architecture that they wanted to incorporate into the data center.
Then we went to the community, and we said, "Guys, we need more land." So they assisted us in procuring 122 acres of land driver nine iron from where our existing site was. They also assisted us in the power pathway to pull the power over with the right of ways and the easements. That enabled us to go to this 20-year triple net, $6.6 billion lease with a high investment grade tenant. So what was important to us, and a lot of really valid points talking about owning GPUs and doing GPU-as-a-service or spinning up your own cloud. For us at CleanSpark, it was about stability. So getting our first lease, $6.6 billion, which equates to $330 million in annual revenue, and being triple net, which means that $330 million annually is almost 100% of the bottom line.
Gives us the stability and flexibility to move into other opportunities that we see fit. Importantly, as we represented the relationship with the community to our tenant, they were impressed with the flexibility that the community showed and their willingness to assist us. So that same tenant asked to get exclusivity on certain other components of our portfolio. So they've also entered into a short duration exclusivity on up to 885 additional megawatts in Texas. Important to note that of that 885 MW, 585 MW was recently announced to be batch zero base load. So the timelines for energization and the ability to deploy a data center there, meets the expectations of our tenant. From a standpoint of construction, I've listened to a lot of my peers talk about the risk, and that's a very real thing.
For us, our tenant actually came with a specific request for an EPCM, and that EPCM has built for them before, and they also manufacture a lot of the mechanical, electrical, and plumbing components internally. It enabled us to get certainty on supply chain. It enabled us to push some of the potential risks and liquidated damages over to the construction site.
We entered into a lump sum turnkey contract that contemplates the first data hall will be delivered in December of 2027. With that, I think one of the things that's important to note, and this was brought up as we've investigated the financing of this lease long term, and that is, in the beginning, the analysis for high yield or investment grade was, who's the guarantor? What's the credit wrap? Who's going to give certainty that the rent payments are being made?
That evolved to, who's the guarantor and what's the certainty on construction? It's now gone beyond that to, who's the guarantor, what's the certainty on construction, and what do the political headwinds look like? We've really done everything in our power to mitigate that risk. We've got a guarantee from the parent, a high investment grade of which there are a handful in the world. Obviously, that'll become public as the financing is completed. The ability to build that quickly has been de-risked largely because of the fact of that lump sum turnkey contract at the request of our investor. The last point about political headwinds, as we've begun to grow the portfolio, we reached out to some of the communities that we operate in, and we've asked them for letters of reference.
To give you a specific example, Jimmy Andrews, the Mayor of Sandersville, wrote a letter of recommendation for CleanSpark for other cities that we may operate in, talking about the community partnership and the win-win spirit that we've operated within those jurisdictions. It really has knocked down a number of the hurdles or the barriers to entry, and we feel very confident that the exclusivity period and the option for that will ultimately be executed, and we're seeing price go up in parallel with demand.
Thanks for that, Matt. Fred, we'll go on to you. MARA has guided to two leases by year-end, and we'd love to get your thoughts on that.
We don't announce LOIs or exclusivities around leases, but we feel very confident about signing the two leases by year-end. I think if you look across our portfolio, the other part of your question was, where do we expect to see stuff happening first? I think you're going to see it both in the small and the large sites, because there's a wide appetite for sites today. Hyperscalers, frontier model providers, neoclouds. There are different size requirements. In some cases, it's 400-500 MW. In other cases, it's 50-150 MW. You can optimize your pricing model to the credit profile of the prospective tenant, to the duration of the lease, to the use case for the site, really to optimize the portfolio value.
You look at a large site like the Matagorda site or the Rockdale site, where you have the opportunity to scale significantly north of 500 MW. You have the ability to have multiple tenants and optimize the value of the portfolio. So if you have a hyperscaler at a triple-net lease where you're talking about a yield on cost of 10%, you may have a neocloud at 15% yield on cost, then you may have enterprise clients at an even higher level. So building a portfolio that maximizes the profitability of the electrons is kind of very much our focus.
John, Soluna is taking an interesting angle with renewable energy assets. If you could share where you guys are on your, I believe, LOI most recently.
Yeah, just for the benefit of the room, if folks don't know about Soluna, we do something unique in that we go to existing renewable energy power plants, both wind and solar, that already have substations that are interconnected to the grid. But they have a problem that they need to solve, and that is about half of their power never makes it to the grid. So they have power that needs to be monetized. They don't have a way to monetize that energy.
We bring the monetization or the load to that wind and solar plant, and then we build a data center campus in that location and go vertical there. For the last five years, we've been doing this in the U.S. We have over 200 MW of that spinning. On the Bitcoin side, we've built data centers that are then leased by very large mining companies. In the last year, we have been transforming the company, taking the same model where we interconnect to the power plant, get access to its power.
Power from the grid, and we can also build on-site generation to build these unique campuses, and essentially have been building out a new series of facilities that will be AI-focused. This model has allowed us to amass over 6.3 GW of power assets over 30 projects. We have about 1.6 GW of that that we are actively developing. There are two projects that we have announced most recently.
One is called Katy II. It is in the southeast part of Texas. It will be a 350 MW campus. We are actively designing and preparing to start construction on the first 100 MW of that, and we did sign an LOI this year and negotiating a lease for that. So that will be our first announced lease for that campus. The second one is Project Dorothy 3, the northwest part of Texas, and that sits across the street from an existing Bitcoin farm.
It also sits across the street from a wind farm that we acquired this year. So we took one step further in our thesis that power assets and compute assets will become, over time, one and the same thing. So Soluna Holdings is building a power compute integrated platform to support the future of AI. So Project Dorothy 3 will be our second campus. We are building that to be about 300 MW.
Part of the power will come from moving the power from the Bitcoin site, and then the rest of the power for both D3, Project Dorothy 3, and Katy II, will come from something we call clustering. Because our pipeline is so big, over 6 GW now, in each site, we have multiple wind farms or solar projects that we can source power from that are within five to eight miles of each other.
We can pull power into a campus and build a very large footprint. Katy II, for example, will benefit from clustering to get it over that 350 MW structure, and Project Dorothy 3 as well will use clustering to bring in over 300 MW of critical IT. So we are in the midst of our transformation. We are very excited about what we have done thus far. There has been some discussion about the risks in building out these new technologies. We focus on infrastructure and power because that is our core expertise, and we are doubling down on that capability, and we are adding to our existing strength, execution capability.
We've hired an incredible leader from Microsoft who will lead up our construction and operations and bringing over part of his team as well that will help us to execute on building our first data center and expand from there and build an entire series of data centers in the AI space. We are marketing these to hyperscale and neoclouds, and the goal is to get the best deal in the first one, continue to double down on those deal structures, and then take that to the capital formation markets as well.
Thank you, John. Josh, over to you on Big Digital.
Big Digital is a relative new player in the space. April 6th this year, my partners, Phil Stanley, our CEO, Cody Smith, our COO, had the opportunity to take control of the former Mawsson Infrastructure company. Since April 6th, we've rebranded the company, cleaned out the deck, and started building as fast as we could. In July, we announced our first acquisition in Texas for the company and our first LOI on an initial 17 MW AI data center build-out. It's been a fast and furious pace for us getting this thing turned in the right direction. At the same time, trying to keep pace with the amount of demand that we're seeing from both hyperscalers and neoclouds for the rest of the assets we have in the bucket.
I think we're most excited about this Hood County opportunity because while it represents a small footprint on phase one, we have phase two right behind it that goes to 74 MW, eventually taps out at about 111. But we recently got confirmation from our energy partner that by way of our two 12-inch nat gas lines and one 20-inch nat gas line on our property, we have a minimum of 600 MW of nat gas generation available to us right on our site.
We look forward to working through the process of doing this dual feed system. Speaking just a little bit about our partner that we chose in TensorIQ, everybody's chasing the IG credit, right? Everybody's chasing the hyperscaler and trying to do the big deals, and I absolutely applaud my peers for knocking those down and out. Our focus right now is on what we're going to look like over the next two, three, four, five years, and what kind of player do we really want to be in this space.
We decided to partner with TensorIQ because their founders bring a collective 40 years of experience and expertise in the data center field, design and construction of data centers, and being able to monetize the GPU end-user space as well. I see us using that relationship beyond just Hood County to our other sites. I think we've got a fun road ahead with them, and at the same time, we're still fielding a lot of requests. Demand has been nonstop since we took over for every available megawatt that we have to us, and I don't see that letting up.
Demand has been nonstop. You hear that, folks? Michael, on to you for Bitdeer.
I'll just, again, touching on the two main businesses we're doing. For the co-location deal we did in Norway, we needed to make sure that this was financeable, and it's important to have a good backstop for it. A lot of people have been using guarantee letters or some form of very strong take or pay lease with investment grade lessor to do it. We didn't have really a great access to either of those. What we ended up doing is having an LC backstop that our customer's providing. This is not something that we have to pay for. It's part of the arrangement we made to lease the site. Now if there's a problem and we don't get paid the lease money, we can pull down the LCs and get cash right away.
It's not persecuting through an agreement or a guarantee, but this is actually an LC that'll turn into cash right away, both from investment grade banks. We think that'll help us in the financing, and we think it's a very solid way to backstop the lease there. The other very interesting part of us doing the GPU-as-a-service in Malaysia is our ability to access very large customer prepayments. Although our customer are investment grade or near investment grade, about 50% or more of the GPU cost is being advanced to us as an advance on the service payments they're going to be making. That greatly reduces the need to finance by 50% or more. It makes it much easier to approach the market and get the remaining financing.
That allows us to scale more rapidly using less capital and less of our own equity in order to spin these businesses up. It turns into revenue fairly quickly for us as well. We're putting all of these into leased sites, so places that we're leasing from another provider in Malaysia. We're able to very quickly execute on these early next year into mid of next year. It's $800 million in more revenue just for the first 10 MW over five years. For us, who's doing about 75 annual run rate of AI business today, which is quite small, rapidly we're doubling and even more quickly when we go from 10 to 20 to 65 MW. We're able to scale that quickly because of our ability to get these large prepayments.
I feel very comfortable that we're going to be delivering really good revenue and profitability next year because of these businesses that are finally turning on. It's the hardest part of the cycle right now for a lot of us. We're spending a lot of money on CapEx, but the revenue isn't here yet. Finally, at this point, three months or so, we'll finally start to be getting revenue for us. That I think will make a big difference in the way that the company is looked at and the way the business opportunities are looked at.
Thank you, Michael. On to you, Russ.
Your first question was on second deal. The deal we just announced about six weeks ago was with AMD. It was for about 500 MW. Your second question was what's our next deals? I'll go and tell you. I'll be very specific. When we signed that 500 MW with AMD, we agreed to give them the next 2,000 MW at three of our sites in Hunt County, Texas, Pecos, Texas, and Muskogee, Oklahoma. We control our own destiny on that 2,000 MW and on those three sites. I'd say unless the market just really goes south, AMD will take all 2,000 of those megawatts. They've already announced who their customers are, so you guys can figure it out. Our next deal will be those 2,000 MW of critical IT load on those sites that we own. The power's in place.
I feel very confident there. I will say, the economics are very different. The CoreWeave deal we had, I do not know, about $5 billion in NRC, and it is $100 MRC, meaning CoreWeave put up a lot of the capital for building out those sites. For this deal, it is a triple net lease at 125, or it is a modified gross at 145 MRC, for 15 years with AMD, which is where the market is right now.
Our deals after that, we have not announced anything yet. Like Fred, I think we are going to have some smaller deals coming as well. I would see not only the very large campuses, but I also see small inference sites being in high demand right now, especially from your labs that just need some of that inference load. That is where we are at right now.
Thank you, Russell. Sam, WhiteFiber has had a very busy summer. It feels like the last few months you have announced an accelerated number of cloud services deals. Would you like to talk about those?
Yeah. We just announced about $500 million of cloud service deals, which is actually even surprising for me. Our cloud service team has been really at it, so we are really proud of that, and we see a lot of stuff in the pipeline as well on the cloud side. That has been surprising for management, how hardworking our cloud team has been producing results.
But just going back on the colocation side, I think it is always very tricky to build and spend all this CapEx on a facility without a contract. That is something we try not to do. We have now six facilities, if we could speak of. The first three facilities in Quebec, Montreal one Montreal two, and Montreal three. Let me just give some color on that. Montreal one we have 20 customers in that smaller facility.
The second facility was a pharmaceutical capsule production factory, which we took over. The third facility, as I mentioned, was a mattress factory, which we turned into an AI campus for Cerebras that is fully dedicated to Cerebras, and we got that up and running on time and within budget. Cerebras has been a great client of ours, and we look forward to potentially working with them again in the future since we have served them well. Now in the U.S., we have been focusing on North Carolina, and the utility company there is Duke Energy. We have a great relationship with them. The NC 1 campus, the one for 1 million sq ft, that is just online, is just basically we are only halfway there with 50 megawatts, and I am rounding some numbers up right now.
Duke has. There is a path towards 99 megawatts associated with that facility. Over the next few years, potentially another 100 MW in NC1. We have the first 50 MW online for NC1 for Enscale. We started billing Enscale, which is great news for us. There is tranche two for NC1, which is another 50 MW. That is something in which we have a very pregnant pipeline of customer interest, particularly investment grade, looking to get that second tranche. They have seen what we have done with Cerebras, they have seen what we have done with Enscale. They like the execution story, and they are looking at the second tranche of NC1 in order to execute for them and get that up and running.
As you mentioned, you did not mention this, but we just put deposits on NC2 and NC3, which is about 60 MW. Again, because of the reputation we have acquired on executing on time, we have investment grade clients who are looking to. Basically, they want us to build it out for them on NC2 and NC3. These are good problems to have, but I do sympathize with the heavy CapEx and trying to match that in time with your contract revenues. That is always a very stressful period.
Multi-billion dollar deals secured, and it sounds like several billion on the way with this group here. You did touch on this, Sam, and I want to go a little deeper onto this. CapEx intensity of AI data center builds is certainly quite hefty. There is no secret there. John, maybe you can discuss the sources of funds, cost of funds, and the cost per megawatt to build out these data centers.
Yeah, sure. Our model is probably no different than others on the panel. We look at a project-level financing that will be primarily debt, 80% of the cost being debt, high yield, attached to the asset, and the remaining 20% through equity, most likely through the topco, pushed down into the project. To give you a sense, just on a theoretical project, call it 100 MW, call it $1.2 billion-$1.4 billion to build, so somewhere about $10 million-$12 million per megawatt.
This depends on where you are building it. We tend to be in remote locations, so cost is higher. Russell's point about labor cost and access to certain critical folks raises the cost to some extent. When you do that 80/20 split, you are talking about a few hundred million dollars of equity that has to be invested in our first, let us say, 100 MW.
But when you look at the returns on that, you are talking about upwards of $100 million to $150 million net NOI. I think we have a slide in one of our investor presentations that teases this out. The yield on cost is going to be in the high teens, and at some point, the cap rate is in the single digits. You get somewhere on 100 MW, $2.4 billion of asset value added to the company. You compare that to our current valuation and asset base, the 6.3 GW that we have is a very valuable asset if we can convert those assets to spinning data centers. Our average construction time is going to be much like anyone else here. You can map that out to what that looks like from a revenue and growth perspective for the company.
What I like to do sometimes for fun is take that framework and then apply it to the 1.6 that we have got spinning and then apply it to the 6.3 to see what the potential value is. The rest is really about execution, making sure you put together a bankable lease, can take it to market so the high yield folks like what they are going to get and feel secure in the asset. Then, of course, you need to execute and actually deliver the data center.
Thank you, John. Anyone else have anything to add on CapEx or funding for these projects?
I agree with everything that everybody has talked about, and one of the constraints, obviously, is labor. A lot of these data center builds, the GCs use travelers, so guys that live in North Dakota that are spending 10, 12 months in West Texas to build the data center. We took a bit of a differentiated approach with our EPCM, and the fact that they build the vast majority of the data center in a factory. It becomes an assembly line process rather than a bespoke one-off deal where you have got a number of different trades trying to work together. That has simplified the delivery for us, but it is also with a fixed price lump sum contract, given a certainty.
So the way we talked about it is, our expectation is that 10, 11 million a MW complete RFS on the data center. We expect, similar to what John's talking about, to finance that at the project level through the debt markets. We're targeting right now high yield investment type or excuse me, high yield investment bond opportunity. But CleanSpark has. We've used our Bitcoin as the means to prevent any equity dilution on the company for that build. So we went into the project, signed the lease, and the very first day, we did our equity contribution. So it is fully funded from our side of the table. Having that certainty of the space in the assembly line and the certainty on the supply chain really de-risks that.
So in our conversations with lenders, that's really what they're focusing on, is what are the potential risks that get in the way? As you think about CapEx and you think about the finance on these, what you also have to consider is the liquidated damages if you fail to deliver on time. That becomes an existential risk for some folks in this business. Because if you make promises and there are GPU purchases that can be many times the price of the data center build, now you've got some real exposure for liability. So getting the certainty there and being able to laser focus in on what that build cost and the timeframe looks like is a differentiator when it comes to financing these projects.
Great insight, Matt. Anyone else have anything to add here?
I'll just pile on a little bit. I think similar to Matt's concerns regarding risk of delivery, it was one of the reasons why we chose to partner with Starwood. We have such a big portfolio that to do this ourselves, especially not being an expert in building data centers for hyperscaler demands, we wanted to increase certainty and reduce risk. Starwood has built over 7 GW of data center capacity for hyperscalers. They have a captive EPCM within them, Worldwide Mission Critical, who's built a lot of these sites. What we're finding is because of the structure of our model, our cash-on-cash returns are significantly greater than the traditional model because we leverage our site as a contribution on the equity stack.
If you use an 80/20 model, the amount of additional capital we have to include over and above the value of our site in these joint ventures is fairly de minimis, which means that the cash-on-cash return is much greater. It also means we can do more sites in parallel simultaneously in greater capacity without it having to all be credit provided by us. We feel very optimistic about that. It is going to allow us to move very quickly as we really get started here, and we are super excited about it.
Yeah. If I may add, we do something similar. We have access to power infrastructure. In some cases, we will have power plants. That land, once it gets fully developed and you are ready to go vertical, becomes much more valuable than what you paid for it, and that can be used as a contribution on the equity side. On the execution side, being a smaller Bitcoin infrastructure company, we were thinking about the risks with execution around these very complex data center projects, and we approached it in two ways. One is, on our initial project, we partnered with a development company that is a group of ex-AWS folks who have experience developing, constructing, and operating these facilities. That has allowed us to design and develop and market these sites in a way that customers would expect.
Then, this past quarter, we started building our own in-house capability, hiring the first major role to the company, a gentleman that has run and built over $10 billion worth of data center projects for Microsoft. I think he built their largest AI facility, and hiring his team here over the last couple of months. So building out our own execution capability to deliver these sites to the end user customers at the level of quality and timing that they expect.
All right. Let us move on to a controversial topic to spice things up a little bit. Some headlines out there would suggest that we are in an AI infrastructure bubble. I would be curious to get the group's thoughts on this, and I will leave it open for whoever would like to start here.
I think if you talk to the people who have a need for the compute, they would tell you there's no bubble. They just need a lot of compute, and they don't have the ability to deliver on it. I think there are two things going on. One is there a bubble in the hyperscalers and this trillion-dollar spend, or is there a bubble in the AI compute world? Two very different things you're looking at. AT&T recently announced that over 40% of their workloads are being done on open-weight, open-source models. They are not using frontier model providers for that. They expect that number to grow to 70% over the next two years. One of the biggest challenges today is cost per token, and about 60% of workloads done by AI today are fairly mundane.
They can be done by an open-source model or an open-weight model on de minimis amounts of infrastructure. There is a huge need for inference. The question is how much of it will happen in the hyperscaler cloud and with the frontier model provider versus how much of it will happen in the enterprise and with the, let's call them neoclouds for now. But for the neoclouds providing a full stack of software services that allow people to run and load whatever models they want to run in fully air-gapped environments. One of the reasons we went after the Exaion opportunity was we wanted to have the capability to run fully air-gapped, fully sovereign inference at scale for enterprises.
Because if you look at what it costs to run an AWS today with dedicated hardware, dedicated resources, versus running in private cloud, it is a 10x higher cost. Companies are looking at AI as a cost item, right? It has become part of their OpEx budget. If they are going to pay huge fees to the frontier model providers versus training their own models and paying once for a model and then running it proprietarily, you're going to see a lot more inference move towards either behind the firewall or near prem, on-prem inference providers who can allow people to run sovereign models.
I think that is going to be by far the biggest thing moving forward in this industry, and you'll still see a lot of revenue going to the frontier model providers, a lot of revenue still going to the cloud providers. The first inkling of this, I think, is how Google reorganized their AI business. Who is it that runs Google's AI business today? It is the guy that operates their cloud business. It is not their AI team.
Yeah. To echo Fred's comments, from CleanSpark's perspective, we certainly don't see it as a bubble. Russ and I had breakfast this morning. We were talking about the increase in rates for these long-term leases and how rapidly they're appreciating. What we really see is that there are three constraints. Capital
Money is flowing where there's the least amount of risk and the highest credit certainty. Energy availability, those energized megawatts, to have conversations with hyperscalers that they're actually contemplating behind-the-meter investments in building their own power plants in order to support the data center, the need for compute that they have, gives you a pretty strong indication that this is certainly not a bubble.
The last thing is delivery certainty. As we've entered the conversations for credit assessment of the project and working with the high yield market, they talk about the fact that certainty of execution is now the predominant factor in determining not only the availability and the willingness for the high yield market to provide that credit, but also drives the rates in a very strong way.
I see that if you have energized megawatts of power and you have those exist in areas that don't have strong NIMBYism. Let's be honest, if we sat here a year ago, I don't think anybody on this panel would have assumed that the governor of Texas would come out and put a moratorium on data centers for certainty of execution. This has become a political hot mess, and both sides of the aisle are mutually opposed to it.
But now we find ourselves with the labor unions in agreement with President Trump that it's an absolute necessity to build these sites here for two reasons. Number one, just like with Bitcoin mining, if you drive Bitcoin mining offshore, it will continue. It'll just continue somewhere else, and that doesn't give certainty on block space for U.S. investors. Number two, these models that are being trained are being trained outside of the U.S., and so that presents an additional level of risk.
We really feel strongly that bringing the right package and the right assets in the right communities and doing an educational process in that is going to be a differentiator to solve for the bubble. In Texas, we worked with our EPCM provider for our Georgia site, and we did a roadshow. We set it up like a trade show with booths addressing everything from dark sky issues and water usage to ambient noise, and impacts on utility rates. If you can lead with education and facts rather than just hype and theory, it's a huge differentiator that we believe solves for a lot of that issue. But as to whether or not there's a bubble, hard no, from my perspective.
I'll add a little bit.
Russ, I have a contrarian perspective, and I think this is a better way to look at the question. I would say that whether it's a bubble or not is kind of a moot point because if you look at technology development and infrastructure build-out for the last 50 years, every single major improvement in our capabilities as a humanity, businesses, technology services on a global basis, has been preceded by major investments like this. The last one was 30 years ago, where people were scratching their head, "What is this Internet thing?" Let me tell you what happened. We built out the most advanced telecom infrastructure capability around the globe. We're all super connected.
We then added to that mobile and content platforms and social connectivity software, which eventually delivered to us what everybody's itching and looking at in their hand, where you essentially have a compute content bundle that gets delivered to you effectively for free. That took billions of dollars of build-out to deliver that type of capability. AI is another one of those. It is the greatest general-purpose technology we've developed in the last 30 years.
I can tell you, I've been a technologist for that time, and I was trained to see these types of waves. This is another big tech wave. You can tell it's a major tech wave when you have infrastructure investments that hit the core elements of the base infrastructure in any given epoch, if you will. This is the first time we're touching the energy infrastructure to power this technology.
Like most of my colleagues are talking about, hyperscalers are looking to build out energy. They're looking to solve the pain, which is speed to power, if you will. That's because the ultimate new bundle is going to be intelligence. We will now have the ability to have intelligence walk around in our pockets at the greatest level we've ever seen, and that's going to be made possible by multiple hundreds of billions of dollars per country.
That's trillions of dollars on a global basis that will be built out to deliver this technology. We cannot stop. As a country, if we stop doing this, we will fall behind because it's so powerful a technology that it's now almost at a military level. Think World War II, where we built out roads to protect the country in that sense. That's the power of this technology.
The reason the companies on stage here are so valuable is because we are part of helping to create that new technology infrastructure. Bubble or no bubble, you can only see it in hindsight, connect the dots, but I am telling you the dots point to a new form of infrastructure that integrates energy, compute, connectivity, and telecom into one bundle to deliver tokens, or otherwise known as intelligence framework or intelligence infrastructure.
There you have it. We are not in an AI infrastructure bubble, at least according to those on set. Sam, did you have something to add here?
Yeah. Just very briefly, I just obviously agree with that. I would say that there are three catalysts for the Industrial Revolution. The first one was electricity, which changed industry, then the internet, which changed industry, and now AI, which is going to change industry. So there is clearly no bubble. But just with respect to the over there is a lot of pushback on data center development, and I think there are many communities these days who are upset that some of us are perhaps taking farmland or building these unsightly data centers in their communities where it just used to be greenfield. That is one of the main reasons why WhiteFiber is retrofitting. We are taking over basically facilities that were left for dead. We have community days.
For example, in North Carolina, we had a community day where we explained that we are using 95% less water than the facility that was being used before we took it over, that we are creating a lot less noise than the facility that we are taking it over from. This retrofit format really solves a lot of the tension that is happening for, against which I think is probably not a great idea. Doing greenfields are kind of you are going to get community pushback, especially if you are taking up farmland. This retrofit format is a way to solve from that community pushback that we have been able to solve with our community day, with the local communities where we are building these retrofitted data centers.
Mike, could I add a quick note?
Sure. Quick.
I wanted to add on to something that Fred touched on earlier about the enterprise client and with the question being, are we in a bubble? The answer for Big Digital is no as well. I spent 20 years in healthcare, just one small sector of this great economy we have in the U.S. And in the last two months, I've had the opportunity to sit down with two founding CEOs of healthcare-focused and drill down to long-term care-focused software companies building AI models to serve that sector.
I heard the same thing from both of them. We have the product, we have the client. Our biggest fear as we scale up and out is that we can't secure the level of compute that we'll need going forward. If I could leave anything with the investors in this room is, we're always thinking about the big five, right?
We're not thinking about the tens of thousands of enterprise-level companies out there that need the same level of compute, and they need it secured and available so that they can grow their businesses. That's something that we at Big Digital are very focused on is the enterprise-level client. More specifically with my background in healthcare, I'm very focused on how can I serve the healthcare client. I would say that the current narrative, whomever it was made up by, I'm sure they had their agenda, but it's completely false.
Let me tell you, there is a bubble. Let me tell you where it's at. There's about 5 GW of rack space being delivered every year, and there's about 15 GW of chip demand every year. But in Texas, what the governor did was reasonable because there was 474 GW of request in for data centers. We don't have enough foundry, silicon foundries on the planet to use that much space. That's what's scaring everybody, is because there's five times the request in Texas that the whole grid commits to. The issue is not the guys delivering rack space or the guys delivering the chips.
It's every farmer and rancher in Texas or Oklahoma or any state that has a transmission line or a gas line running through their site that says, "I'm going to put a data center there." We get people telling us, "I've got a 3 GW site." The next guy beside him says, "I got a 3 GW site." The next guy beside him says, "I have a 3 GW site." They're all counting the same gas line, and they're all counting the same gas. That's the issue that the governor's office and ERCOT had in Texas. It was how many of these data center requests are actually real? There really is 15 GW a year of demand, and we really are. We're not able to fulfill it, and it's going to go on for as long as we can.
What there's not is nearly 500 GW of demand in Texas alone. That's where I think I actually don't think it was a bad thing that the governor said, "Hey, wait a minute. How much of this is real? Who has a substation? Who has a PPA? Who has power? Who has your behind-the-meter plan together?" I think it's reasonable for any state to ask that question because all these are not real. We saw the same thing in Bitcoin mining. I know some of you guys were there. Texas ERCOT told us, and by the way, Core Scientific is one of the largest users of power. We talk to ERCOT multiple times a day, all the time. ERCOT told us there was 20 GW of Bitcoin mining demand in Texas. There wasn't. There's like a fifth of that.
What it was the same guys chasing the same. Everyone wanted to sell me some stuff, they want to sell Fred some stuff, they want to sell Matt. It's the same, it's people just brokering these deals or putting deals together. That actually is a bubble. There's not tens of thousands of gigawatts of demand in every single state in the country. I think what needs to happen is some regulations on, hey, what's real and what's not real. I think that's an unbelievably reasonable request for anybody to make.
I agree with that.
Just to level set, it takes five to six years to build a new gas-fired facility. While solar and wind are great, it is intermittent, until we manage the orchestration issue. If the U.S., I mean, Duke did the study last year. The U.S. has 72 GW of power available to it based on current generation today. You want to add anything to that?
It is all. It all takes a couple to six, seven years and billions and billions of dollars to do it.
Yeah, there's a dislocation.
All right, last question. We do have three of the largest publicly traded Bitcoin miners on the stage here. I would be remiss to not ask a question about Bitcoin and mining. Specifically for MARA, CleanSpark, and Bitdeer, if either of you had something to add, specifically, how do you plan to leverage your mining assets and Bitcoin holdings to really fit into your longer-term strategy as you continue to execute on the AI data center side?
Well, Bitcoin was obviously higher a year ago, and it was obviously lower a month ago. We've developed an institutional-grade trading desk to manage our Bitcoin. If you go to BitcoinTreasuries.net, I think we're eight or nine on that list. Fred, I think, is two.
Not anymore.
But we have over $1 billion in Bitcoin, and we have $400 million in largely untapped Bitcoin-backed lines of credit, and we have an institutional-grade trading desk to help generate yield on that $1 billion plus of Bitcoin that we hold. We see it really as a capital tool. Additionally, we've prevailed on land and power sites because of Bitcoin. In Cheyenne, Wyoming, as an example, 110 MW from Black Hills Energy, they put out an RFP to acquire that, and there were two bidders. One was a trillion-dollar hyperscaler, and the other was CleanSpark. And we prevailed, not because we have a better balance sheet or we're more handsome, but because of the fact that we were able to deploy immersion-cooled Bitcoin mining infrastructure in a modular basis and start paying a power bill in 90 days.
That allowed us to win a contract over a trillion-dollar company. Now, we bought enough land that we can build a proper data center to monetize those megawatts when we have the right client, and we can build to their reference architecture and their basis of design. But to use Bitcoin mining to profitably monetize those megawatts now and to secure the contract, we believe, is a significant differentiator.
Yeah. I totally agree with Matt. Being a Bitcoin miner, you have a load you can deploy immediately. If you are in a queue for load and you're waiting for energization, you've got something you can turn on right away, even before the AI data center is built. And our whole model works on essentially converting Bitcoin mining sites to AI. It allows us to move as the tenant starts moving into the data center on the AI side, we can start deprecating our Bitcoin mining, moving it to another site that we acquired, and continuing to push that down the line.
Bitcoin miners have an advantage today, structurally, in that the cost of Bitcoin mining ASICs has never been as low on a per terahash basis as it is today. We all have incentives to optimize the existing mining we have at very low CapEx rates, which allows us to, as our megawatt capacity allocated to Bitcoin shrinks, our exahash capacity doesn't have to. I think that's an important benefit that we as miners all have.
Yeah. I think Bitdeer is the largest public Bitcoin mining company right now. We make our own ASICs. We think they're great ASICs, and they're very efficient. We generate more cash than the cost when we're doing Bitcoin mining. As Ben mentioned, we can do Bitcoin mining all the way up till you turn the data center on, and if there's a corner of a land and a little bit of extra megawatts that's not being used for the data center, you can turn it into cash by the Bitcoin mining.
We're quite comfortable with our current plans. We are slowly expanding our hash rate. I wouldn't say slowly, but we're not investing a huge amount of money and time and effort into doing it, but it is going up. We're continuing to design new Bitcoin mining ASICs, and we're excited about the possibilities for that as well. We're happy with what we're doing, but most of our focus today is on the AI data center build, and it'll continue to be on that area.
I like to say one of the other topics that we're talking about, this whole bubble and such, this is only because we're in the massive investment phase before the revenue turns on. Because there's lots of money going in being spent, but no revenue to go with it worth mentioning right now, people see the spending without the revenue, and the bubble worry always bubbles up. For us, starting next year, just the Malaysia pipeline we talked about, it's close to $7 billion or more revenue over five years. We have almost no Bitcoin. It's not. We have almost no AI data center revenue today. Suddenly, that's over $1 billion a year worth of revenue.
When the revenue shows up, it won't really be a bubble conversation. It'll be, "Why aren't you doing more faster?" I think that'll counterbalance that. When some cash flow starts happening and profit starts happening, people will be very comfortable. I think Bitdeer, combined with the foundation of the Bitcoin mining we've done for a while, has given us the power and land, and it lets us exploit it to the best advantage we can. Then we're going to start delivering really good results on the AI data center. I'm quite comfortable talking about Bitcoin when it comes up. I spend almost all my time on AI data centers.
From our perspective at Soluna in this topic, Bitcoin is a core part of our business. We are not one of the largest Bitcoin miners. We are not a Bitcoin miner at all. We are one of the largest Bitcoin hosting companies that hosts some of the largest miners. That business approach allows us massive amounts of flexibility, especially as it relates to partnering with these large power plant owners. They have a problem. They have wasted energy that they need to monetize very quickly to return these projects to profitability. We can deploy Bitcoin mining to some of these sites, especially the smaller ones.
We can continue to grow our hosting business, build a relationship with these power asset companies, which allows us to then monetize that asset while we are developing the AI project, which tends to take longer, and also show them a quick win for monetizing that power asset, which unlocks for us a lot more projects that are even bigger. That is how our pipeline has grown so significantly over the last few quarters, because we have developed this very strategic approach to monetizing their power fast and then opening up the door to a longer relationship where we can build much larger scaled operations on the AI side. Bitcoin has become a very strategic platform for us as we transition into AI.
All right. With that will conclude our panel discussion for today. Thank you all for attending.