Marathon Digital at KeyBanc Forum: Strategic Expansion into AI and HPC

Marathon Digital Holdings (NASDAQ:MARA) took center stage at the KeyBanc Capital Markets Technology Leadership Forum on Tuesday, 12 August 2025. The company, renowned as the largest publicly traded Bitcoin miner, unveiled its strategic pivot towards AI and high-performance computing (HPC). While highlighting operational efficiencies and future growth prospects, the discussion also addressed challenges in the evolving competitive landscape.

Key Takeaways

• Marathon Digital is expanding into AI and HPC, starting with a control investment in a French technology company.
• The company operates 16 data centers across four continents, emphasizing renewable energy and operational efficiency.
• Marathon aims for a balanced revenue split between U.S. and international operations within five years, with AI playing a significant role.
• The firm holds the second-largest publicly traded Bitcoin treasury, actively managing it to generate yield.
• Marathon’s strategic focus includes a recurring revenue model in AI to counter the cyclical nature of Bitcoin mining.

Financial Results

• Electricity cost per coin was under $30,000 last year, with an electricity cost of 4¢ per kilowatt hour on owned power generation.
• Marathon raised $2 billion with a 0% coupon last year and an additional $1 billion recently.
• The company has over $5 billion in treasury, earmarked for AI or Bitcoin investments.
• Return on capital employed is reported to be three times better than the nearest competitor.

Operational Updates

• Marathon increased its owned and operated sites from 0% to 70% by the end of last year.
• Investments in renewable energy include a wind farm, and operations in Finland contribute to heating 80,000 homes with data center heat.
• The company operates four Tier 3 and Tier 4 data centers focused on HPC and AI.

Future Outlook

• Marathon targets a 50/50 revenue split between U.S. and international operations.
• The growth of sovereign AI is expected to be substantial, with the inference as a service market projected to reach $160 billion by 2028.
• The company plans to leverage its technology stack and customer relationships to scale its business.

Q&A Highlights

• Industry consolidation is expected to continue, with power concentrated among a few large players.
• New entrants are driven by decentralization and censorship resistance.
• Marathon’s competitive advantage lies in its technology and infrastructure investments, leading to operational efficiency.

For a comprehensive understanding of Marathon Digital’s strategic direction, readers are encouraged to refer to the full transcript below.

Full transcript - KeyBanc Capital Markets Technology Leadership Forum:

Unidentified speaker, Moderator: Everyone for for being here. Excited to have the Mara team with us. We have Fred Thiel and Salman Khan. We’ve got twenty five minutes, so I’ve got some Q and A we’ll run through, and we’ll go a bit deeper than the corporate panel, if any of you were in here for that. And we’ll open it up for questions at the end.

I assume that either folks remember what you said an hour ago or are familiar with you, but maybe just thirty seconds on Mara, just in case there’s anyone new in the room.

Fred Thiel, Mara: Sure. So Mara is the largest publicly traded Bitcoin miner and second largest publicly traded of bitcoin in the world. We operate data centers on 16 data centers on four continents. And we also owned wind farms. We own flare gas power generation sites, and we use energy off the grid.

We have operations and joint ventures in UAE with the sovereign there. We operate in Finland, for example, where we heat 80,000 homes with the heat from our data centers. And we also operate in Paraguay. The majority of our operations are in The U. S.

Are technologically vertically integrated, everything from ASICs all the way up to orchestration software. And we recently, as of yesterday, announced an investment a control investment in a French technology company, which is a leading provider of inference at the edge, sovereign pool sovereign cloud, and ESG controlled compute, if you want to call it that. It’s a company that was originally owned by the founders and the corporate venture arm of EDF, which is the largest low carbon energy company in the world. EDF remains an investor in the company, and we now control a little over 60% of the company with an opportunity to grow our ownership even more. And they operate four data centers, all HPC, AI, that are Tier three and Tier four.

So that now allows us to enter the AI HPC space in a very strong way on an international basis.

Unidentified speaker, Moderator: I want to get there. But before we go there, maybe just to sort of level set across the core mining business and just the landscape competitively. I mean, I think the scale of MARA is understood. But maybe just paint a picture who you see out there and how it’s sort of evolved over the past several years. And then anything on new entrants that you’d flag?

Fred Thiel, Mara: Sure. So the industry started as a bunch of, you know, like the tech industry, a bunch of garage shops. And then different people tried to grow at different rates. And starting in really about 2018, 2019, all of a sudden, capital started flowing into the space. And so companies like ourselves that were already public doing other things had the opportunity to go raise money in the first kind of real bitcoin waves, and we were able to raise a lot of money in scale.

We happened to choose a path that was asset light originally. So instead of investing in infrastructure and compute, we just invested in compute and rented space. And power that led us scale to be the biggest in the world. Today, if you look at the industry, there are four miners who together produce, control about 20% of all the bitcoin mined in the world, and those are all American companies, publicly traded companies. So ourselves, CleanSpark, for example, Iris Energy and Riot.

As you look at new entrants, overnight, all of a sudden, have people like Eric and Donald Trump Jr. With American Bitcoin who partnered with an existing Bitcoin miner on our company. You have Tether who have announced publicly, Pablo announced publicly, he wants to be the biggest Bitcoin miner in the world by the end of this year. He’s doing a good job buying up every single Bitcoin miner he can. He also wants to be the biggest holder of bitcoin in the world.

But he’s got a deep wallet, so. But so you’re seeing kind of this concentration in the industry, certainly, and we’ll see more and more consolidation over time.

Unidentified speaker, Moderator: It’s kind of an interesting contrast to on the flip side, there have been some wholesale exits from I shouldn’t say exits pivots from miners to HPC. Where is the disconnect there, right? Like what are new entrants seeing that these folks who are pivoting away from mining?

Fred Thiel, Mara: So I think the new entrants are predominantly looking at bitcoin mining as what is the function of Bitcoin mining? It’s the processing of transactions on the Bitcoin blockchain. You assemble transactions, and then you write them into blocks. And if you win the compute race, you get the right to write that block, and then you write it. So you essentially have control of what transactions get processed.

Tether’s whole premise with becoming the biggest Bitcoin miner in the world is so that they can ensure that the US government doesn’t censor the Bitcoin blockchain by forcing all US miners to be OFAC compliant. This is Paulo’s whole model, as he’s always kind of taking this opposite kind of model. He loves decentralization, hates centralization. He’s willing to lose money doing it. So that’s relative to the new entrants.

I think as you look at the pivot, you have miners who have assets where they sit near some place that has a high speed Internet connection, fiber connection. They’re close to a major metro area. And you have all of these large hyperscalers running around saying, I need power, I need power. And they think, oh, I can convert something that was trading at, let’s just say, dollars 1,000,000 a megawatt into something that trades at 10,000,000 a megawatt overnight by just saying it could be an HPC spot. The truth is, of all the people that have pivoted, only really two have been able to announce deals.

Core Scientific, who Core Weave is now in the process of trying to acquire. And I say trying because a lot of Core’s shareholders are now starting to push back on that deal, at least ask for a higher price. Core Weave, have to realize, is just a middleman. All right? They basically rent capacity and space from facilities and then sell that to Microsoft and others after putting equipment in it, kind of like what we did in the very beginning.

They now are vertically integrating. And so then you have a handful of others who have been able to sell small tranches. Galaxy is the only other one. And by the way they also are selling to CoreWeave. So if you were to rip CoreWeave out of the equation nobody has any deals of note.

And when you really look at this, you have to think about the hyperscalers in the world of LLMs. They’re looking for 500 megawatts to a gigawatt of power at a time. There aren’t any miners that have those type of scales other than Riot. They’re the only ones with a site of that potential scale. And that’s their Corsicana site, which is most probably why they have an activist investor and they’re advocating for them to become an AI HPC shop.

But otherwise, you really, to make money in AI, you can only charge for tokens. By this, I mean AI tokens, which are measure of compute output. And you can only do that via inference. Right? Learning models don’t earn money.

It’s the querying of the learning model that generates revenues, and that’s inference. And the vast majority of inference will happen at the edge. It won’t happen in the main cloud. Every factory, every airplane, every thing that needs to have an AI model running it, that has to run next to the thing because there’s network latencies, there’s the risk that the network goes offline for a second, things like that. And so these big hyperscaler clouds are really full of data that the hyperscaler themselves need to operate.

Meta needs to run inference about all their users to know what ads to serve them. Amazon needs to know what products to recommend. Microsoft needs to essentially analyze the corporate data they’re managing such that Copilot works and can recommend. 70% of corporate data still sits behind corporate firewalls. It’s not in the cloud, regardless of what Amazon, Google like to say.

And so inference at the edge is really where the business is going to be.

Unidentified speaker, Moderator: Yeah. Salman, do you want to maybe talk about the model, the mining model today from a financial perspective? Why it’s getting more challenging for some folks? And then just as you think about this HPC opportunity, like how the model for Amara can change over time?

Salman Khan, Mara: Look, mining business is a very unique business. It’s very different from other industries. I spent twenty years in oil and gas industry, and I thought coming into this industry that it’s a commodity based industry and it’s going to be very similar. Yes, there are similarities, but there are very differences huge differences compared to that sector. So when you think about this sector, you are as a producer of bitcoin, we produce bitcoin.

We have to be very efficient not only from a CapEx perspective, return on capital employed, which, by the way, we are three times better than our closest competitor when it comes to that metric. We also have to be highly efficient on the operating side, so both balance sheet and P and L management. And our electricity cost per coin is one of the lowest in the sector. Last year, we were under $30,000 per coin, for example. Electricity cost per coin is one of the most important metric in this sector because that’s the biggest line item in the income statement.

When miners operate, they operate data centers, they consume electricity and convert those to bitcoin that we hold on our balance sheet. So from a modeling perspective, from understanding how the economics work, those are things that are within our control. Now there are many things that are not within our control. One is the global hash rate. The global hash rate directly impacts how much reward do we earn, and that is dependent entirely on how much bitcoin mining is happening worldwide.

The global hash rate has grown over the years. It is expected to grow further. As it grows further, if miners stop growing, their costs will become much more higher to operate than they were previously. So miners have to continue to grow as a result of that to keep up with the growth of the global hash rate. So that’s the other aspect.

Then you also have treasury management, which which is another aspect to which is sort of unique to Mara. We are the second largest holder of Bitcoin, and we’re not a treasury company like MicroStrategy, but we happen to have the second largest bitcoin on our balance sheet. And we like bitcoin. It has given us great rate of returns historically, but we are very prudent in how we manage that. We have an asset management team.

We are constantly looking at optimizing that asset and squeezing more out of the system, not only just capital appreciation, but also a yield on bitcoin, which I’m sure we’ll talk about in a moment. But the industry has its own nuances. It can create significant value over a period of time provided that you have the right mindset and that strategy translates into action. I’ll give you one example really quick. Fred mentioned a few moments ago the history of the company.

We were asset light. We became the largest operator as a result of deploying capital on a rented rack space. The result of that was we were the largest, but our costs to operate was on the higher side because we were paying somebody to operate on our behalf. Last year, we entered the year with 0% owned and operated, which meant that 100% of our operations were through third party operated sites. We exited the year by owning 70% of our production or capacity, whereas 30% is still left, means that we have more opportunities to reduce our cost.

The result of that was that our electricity cost per coin is now one of the lowest in the sector. We we pay 4¢ per kilowatt hour on the power side, and that is one of the lowest in the sector, which is very important from a strategic perspective. No matter what happens on the Bitcoin price, every four years, this halving concept, which is important concept in this sector, where your reward halves compared to what it used to be. So you have to keep up with the cost. You have to continue to reduce cost.

We did not stop there. We reduced our electricity cost, but we also invested, Fred mentioned, in renewable energy resources. We bought a wind farm. Now we’re not an electricity company. We don’t want to be seen an electricity company.

We don’t want to sell electricity either. But we like to consume electrons when they’re producing from a natural resource at almost zero cost. Because when the wind blows, when the sun shines, electricity produced at almost zero cost, marginal cost, and that’s when we produce bitcoin on it. And because of the intermittent nature of bitcoin mining, we don’t have to produce when the wind is not blowing. We shut down the operations at that time and still get the better ROI than a traditional grid connected side, about 85% to 90% better economics than a grid connected side.

So that’s what we’re excited about on the bitcoin mining side. On the HPC side, that’s the we’re still a bitcoin miner primarily, but we also like diversity in revenues. Our KPI primarily is profit per megawatt hour. When I say profit per megawatt hour, that really means cash flow per megawatt hour. Have to think about it from that standpoint.

And with the eye to become free cash flow despite this being a capital intensive business on both sides from a long term perspective.

Unidentified speaker, Moderator: And maybe just to take that last point and run with it a little bit on the HPC side. You talked about the stake announced yesterday. What’s sort of on the how should we think about the road map to realize this opportunity that you’re describing and thinking about the evolution of MARA over a five, ten year horizon?

Fred Thiel, Mara: Right. So the great thing with the data center business, unlike the bitcoin mining business. So bitcoin mining, it’s a zero sum game. There are only so many bitcoin that are awarded for every block and that gets havoc every four years like someone said. So the more compute that you apply to it, you don’t increase the amount of bitcoin you’re going to get if everybody else is increasing pro rata, right?

In the data center world, it’s a land grab. And so if you have a technology solution and a product offering that’s really relevant to a segment, then it’s kind of like being in the dry cleaning business, right? You have a back end dry cleaning production facility and then you just cookie cutter set up dry cleaners front ends. Same thing in the data center business. Inference at the edge truly is at the edge.

And you set up small data centers, small meaning not 500 megawatts, you’re talking five to 10, max 20 megawatts maybe, data centers in lots of locations. And you scale the business. You’re leveraging that same technology stack and the same set of customers even because they have different branches, they have different operations facilities. And you go after sectors like health care, financial services, manufacturing, industries that need data processed on-site, that need sovereign data, they don’t want to let their data up into the cloud, and that have a high need for inference. You know, the inference market according to Gartner or Inference as a Service, Infrastructure as a Service is a $160,000,000,000 market by 2028.

Yeah, it’s a hellacious growth rate. So we believe if we just take a little sliver of that and then if we do it in a place where there’s a lot less competition, there’s a lot less competition in Europe because they want sovereign. They don’t want big U. S. Companies coming in and setting up shop.

They want local companies. So we set up a local MARA of France. We’ve set up a whole process to do that. And we also do the Middle East and The Gulf regions where we’re already strong in UAE. We’ve now set up a regional HQ in Saudi.

And we think those areas are going to continue to invest very heavily in being sovereign clouds. So people don’t put data in Chinese or US clouds. Mhmm. And so we think there’s a that’s a very big opportunity for us to scale. And, you know, done properly, you can use lots of project finance capital.

It’s a whole different model in the Bitcoin world, and Salman will attest to this. You know, up until very recently, you couldn’t get a bank to even open a checking account for you, let alone lend you money to build a facility or buy capital equipment. Totally different story in the AI world. There are people willing to throw money at you in that sector. Mhmm.

Unidentified speaker, Moderator: Yeah. I mean, it seems like there’s this sort of, you know, pivotal point for a lot of miners to to take advantage of a large opportunity, I guess. Salman, when you think about the MARA model in realizing this opportunity, this HPC opportunity, how does it look in five, ten years if successful? And maybe fold in the deployment of Bitcoin on the balance sheet, yield generating activity? Just how should we think about the evolution of maybe those three buckets, mining, deployment of the balance sheet and HPC and just put it all together for us?

Fred Thiel, Mara: That’s a great question. You ask me this question every day.

Salman Khan, Mara: So talking about Mara five years down the line, we would be fifty-fifty split between international and domestic operations. Currently, most of our revenue is coming from domestic U. S. Operations, although we have presence worldwide, but dollar wise. And in terms of how AI is a fast growing sector and that can grow very, very quickly in meaningful size.

That 50% could become interesting and challenging for us in five years, which is a good problem to have. Yet at the same time, I would expect there are so many uncertainties on the pricing and macroeconomics, but I would expect that we will continue to grow responsibly on the mining side and continue to deploy treasury strategies depending on market conditions. So for example, last year, we were beginning of last year, until that, we were selling bitcoin to fund our operations. Just to pause for a second, if you’re a furniture manufacturer and you sell furniture, you want to see cash against that and pay for your expenses. But if you’re getting cash that you receive from the sellers and that gives you a rate of return of 50%, 60%, 70% IRR in a short period of time, you would want to deploy that into that investment as a treasury, return on capital will look much better on that versus reinvesting in the business of manufacturing furniture, which may give you a rate of return of 2010%, 2015% or so in that industry.

So take that example, apply that to us. It’s a dynamic model where we look at whether it makes more sense to invest deploy our capital on buying Bitcoin, which we have bought last year. We raised $2,000,000,000 last year with 0% coupon. We did $1,000,000,000 more than $1,000,000,000 this year, a few weeks ago, seven years out. It’s one of the longest duration a testament by the industry and investors that we’re the closest to the investment grade as it would be in this sector.

Reminder that there’s no credit ratings in this sector as yet. It’s a much newer industry. But it’s a testament of what Mara has been able to do from a balance sheet and P and L perspective. So five years out, I would expect us to be significantly different from now where we are with our strategic plan of growing fifty-fifty split between U. S.

And international with sovereign AI continuing to grow significantly and be a meaningful part of our revenues, while bitcoin mining continues to be a growth engine for us. And treasury depends on opportunistic yield generation. If it generates more to invest in AI or bitcoin, then we would be willing to take that capital that we have sitting on our balance sheet, more than $5,000,000,000 and take that and deploy it to create value for our stockholders. If bitcoin as a treasury asset is the best case, then we’ll stay put and continue to create that yield that we’re working on.

Fred Thiel, Mara: Yeah. I think an important difference in the two business models. In the world of bitcoin mining, we’re essentially being paid a subsidy plus transaction fees. And it’s very capital intensive. We have to continually reinvest.

In the world of what we’re doing in AI, it’s very much a recurring revenue model. One of the reasons we went this route of investing in a company that already has customers, that already has data centers versus hiring a bunch of people and trying to go sell somebody so we could go build one is that the roadblock the Bitcoin miners that have tried to pivot have hit is they have not proven that they know how to do it. So instead by buying somebody who knows how to do it, has the right type of customers, has tier three and tier four data centers, we right away get a stamp of credibility. And we can right ahead go to the scale button as opposed to first figure out what your product is, etcetera. So we believe that because of the recurring revenue model that exists on the AI side, you know, you can just look at SaaS companies and see how they scale once they get traction.

And, you know, if and when we get good traction on that side, it can scale at very high margins. And unlike Bitcoin, you don’t have halvings every four years and you don’t have all those other things.

Unidentified speaker, Moderator: Yeah. Yeah. Good. We have a a maybe a minute left. Any questions from the audience before we keep going?

Fred Thiel, Mara: We’ve wowed them Maybe

Unidentified speaker, Moderator: just to wrap up, Fred, I’d love to get your perspective on I’d like to ask like what do you think is misunderstood or missed or underappreciated about the business? And can take it the mining direction or the HPC direction, whatever you prefer? I think

Fred Thiel, Mara: people think of us as, oh, you’re the biggest bitcoin miner in the world, so you should just focus on the efficiency of your mining business. What they don’t realize is we spend a lot of time and energy mental energy at looking and building these technology stacks that make us differentiated. Owning your own technology all the way from ASICs all the way up to pool or orchestration layer on the AI side all the way down to how you run your data centers drives your operational efficiency. And when you make those investments, they pay off over the long term. And so unlike people who just run data centers or just run bitcoin mining sites, we are a technology company that invests in the infrastructure.

And over time, that technology becomes the moat and the differentiator, the ability to do things that other people can’t at a cost point that other people can’t. Salman mentioned that we have the lowest energy cost per bitcoin amongst our peers in our owned and operated sites. Why can’t third party operated sites generate that level of efficiency? A, they don’t have our tools, they don’t have our technology. And lastly, they don’t have teams that are brutally focused on optimizing.

And I think when we apply that to the AI side, you’re going to see some really magical things.

Unidentified speaker, Moderator: It’s a great place to end. Thank you both for joining us here.

Fred Thiel, Mara: Thank you.

Unidentified speaker, Moderator: Thank you. Thanks everyone.

This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.