Note, this conference is being recorded. I will now turn the conference over to Robert Samuels, VP of Investor Relations. Thank you, Robert. You may begin.
Thank you, operator, and good afternoon, everyone, and welcome to MARA's third quarter fiscal year 2026 earnings call. Thanks for joining us today. With me on today's call are our Chairman and Chief Executive Officer, Fred Thiel, and our Chief Financial Officer, Salman Khan. Today's call includes forward-looking statements, including those about our growth plans, liquidity, and financial performance. These involve risks and uncertainties, and actual results may differ materially. We disclaim any obligation to update these statements except as required by law. For more details, please see the risk factor section of our latest 10-K and other SEC filings. We will also reference non-GAAP financial measures like adjusted EBITDA, which we believe are important indicators of MARA's operating performance because they exclude certain items that we do not believe directly reflect our core operations. Please see our earnings release for reconciliations to the most comparable GAAP measures.
We hope you've had the chance to read our shareholder letter and look forward to your feedback. We'll begin with some prepared remarks from Fred and Salman. After their comments, we will open the call to Q&A. I'm going to turn the call over to Fred to get things started. Fred?
Good afternoon, everyone, and thank you for joining us. The third quarter marked an important shift for MARA. Over the past several years, we've assembled a difficult-to-replicate portfolio of powered assets, strategic partnerships, operating infrastructure, and technical capabilities. We have made good progress continuing the transformation of MARA into a business designed to produce contracted returns and durable free cash flow with the simple goal of maximizing the value of every megawatt we control. The market is becoming more selective about how AI infrastructure is financed and delivered. Tenants, lenders, and investors want more than a position in an interconnection queue. They want near-term power, a suitable site, the right development partner, creditworthy tenants, and a credible path to completion. Those requirements go directly to MARA's strengths, and in the third quarter, we put them to work, expanding our power position, advancing tenant discussions, and strengthening our capital flexibility.
Regarding tenants, we have made great progress, and discussions are in very advanced stages. Our engineering and operations teams have hosted prospective tenants across nearly all of our sites, including Hannibal, adjacent to Long Ridge, and we feel particularly proud of the development team's progress at Kearney, where permitting is well advanced, as has been disclosed in the news. We are still targeting two lease signings before year-end, and although market conditions are fluid, that timing could move into next year. We are continuing to advance power, permitting, site, and financing work required to support those agreements and ensure timely delivery. In regards to power, at Long Ridge, we recently received FERC approval, which was the last remaining material contingency, and we expect to complete the acquisition over the next few weeks, subject to customary closing conditions.
We are excited to add a highly efficient natural gas plant to our portfolio that is currently authorized to sell 485 MW and expected to increase to the full- 505 MW nameplate capacity in the first half of 2027. Long Ridge will not only immediately strengthen our financial profile with contracted cash flows from power generation, but also provides us over 1,600 contiguous acres with access to power, water, fiber, and rail, supporting a premier data center campus.
Together with our existing Bitcoin mining operations at the site, we will have over 1 GW of total potential capacity. We look forward to completing the transaction and beginning the next phase of Long Ridge's development as part of our strategy. We also secured rights to a site in Matagorda County, Texas, with access to up to 2 GW of power capacity that has been classified as study load under ERCOT's Batch Zero process.
Today, we operate approximately 1 GW across U.S. and international markets. Subject to the closing of Long Ridge and the energization of Matagorda, we expect that portfolio to reach over 4 GW. Exaion delivered early commercial proof this quarter, signing a four-year, $79 million GPU-as-a-service agreement with a U.S. customer under 100% take-or-pay terms. Exaion has built a pipeline of over $3 billion in total contract value, with approximately $1 billion of that targeted for delivery by the end of 2027. These are indications of demand and are not yet contracted or committed. However, we view this as an early indication of demand for the broader model Exaion is building as part of MARA's broader strategy. Regarding our balance sheet, after the quarter ended, we sold Bitcoin to repay $750 million of Bitcoin-backed debt, releasing 18,750 Bitcoin that had been pledged as collateral.
Together with the $1.4 billion, or 37% of total debt reduced since the beginning of the year, we have strengthened our balance sheet and our position for our next period of growth. We will continue to manage our Bitcoin holdings, liquidity, and debt obligations against market conditions and the expected return on potential uses of capital, including AI infrastructure opportunities. These milestones matter because investors are paying closer attention to the pace of AI adoption, higher financing costs, and the credit structures behind the individual projects. Those factors affect the timing and economics of development, but they do not change the underlying need for power-ready infrastructure. What we hear from prospective tenants is consistent. Demand remains strong for companies that have a credible path to power and delivery. Since the election, there has been a significant uptick in demand for our sites, which we believe provides opportunities for pricing power.
This is also true for some locations in our portfolio that once would have been viewed as peripheral, including parts of our international portfolio. That demand is important because it validates the platform we have spent years building. Our platform was built through Bitcoin mining. Running a large mining fleet requires power procurement, modular deployment, infrastructure optimization, and real-time load management. Those are the same capabilities required to operate a broader digital infrastructure platform, especially as modular data centers are being considered for innovative solutions that shorten delivery timeframes and reduce build costs, as was seen in the Bitcoin mining industry. We believe MARA is one of the largest operators of modular Bitcoin mining digital infrastructure in the United States, and we intend to apply that experience as we transition towards AI HPC.
Bitcoin remains our operating foundation and a flexible way to monetize power while we develop higher value infrastructure opportunities. That foundation gives us optionality. We can keep each megawatt in Bitcoin mining or direct it to AI or HPC when those uses offer a greater risk-adjusted return. That foundation supports three complementary paths to deploy the platform. Mining generates near-term cash flow and keeps energized sites productive. AI and HPC campuses convert powered land into longer duration contracted infrastructure for hyperscalers and other large compute tenants. Modular data centers can bring compute online faster, where a small or faster deployment makes sense. The right solution will depend on the customer, the site, and the return profile. Conventional and modular solutions can co-exist across the portfolio. Once the first phase of a campus is complete, the hardest work on land, interconnection, permitting, and core infrastructure is behind us.
A tenant may begin with 200 MW and ultimately require 400 MW, 600 MW, or more, allowing us to increase the value of the site without starting over somewhere else. Large campuses can also support a mix of hyperscalers, AI cloud providers, and enterprise workloads, which can improve utilization and reduce dependence on any single deployment. Our proposition is built around speed, certainty, and reliability. Speed comes from controlling sites with power already available or on a credible path to availability. Certainty comes from combining power, land, permitting, development, and financing into a project that can be underwritten. Reliability comes from years of designing, owning, and operating large-scale compute infrastructure together with strategic partners like Starwood. Exaion takes MARA above the infrastructure layer. It provides private cloud infrastructure, open weight model orchestration, and inference as a service for customers who need greater control over their data, infrastructure, and jurisdiction.
Those needs are especially important for governments, regulated industries, and enterprises in Europe and other markets where sovereignty and security shape technological decisions. We are commercializing Vertebrai, our flexible energy compute platform built on years of power-intensive operating experience. It dynamically directs power as supply and demand change, a model we first proved at our wind farm in Hansford County, Texas, where prices can fall below zero and curtailment is common. From April through September 2026, we estimate that Vertebrai helped our Texas wind farm cut curtailment by more than half. Without Vertebrai, an estimated 46% of potential wind generation would have been curtailed. By using on-site computing to absorb power that otherwise would have gone unused, Vertebrai reduced that figure to 22%, putting more of the farm's generation to productive use and creating an additional source of revenue.
With Vertebrai now live, we are focused on bringing this capability to additional power producers and also data center developers. We have been deliberate about the projects we pursue. The first wave of AI infrastructure saw aggressive pricing, rising rates, and uncertain delivery schedules put pressure on project economics. That experience reinforced our focus on contracts that can be financed, built, and operated at acceptable risk-adjusted returns. We are focused on the quality and durability of each opportunity, not growth for its own sake. That discipline also guides capital allocation. We manage cash, Bitcoin, debt, and access to capital as parts of a single balance sheet. Bitcoin can be held as a strategic asset, monetized when the return on another use case is more attractive, or used as collateral when this is the most efficient source of funding.
As Salman will discuss in just a minute, Bitcoin has been a differentiating tool for capital management. After quarter end, our board of directors authorized a stock repurchase program of up to $500 million of our common stock. This does not mean that we will not continue to invest in high-return projects. However, with current market conditions, we believe having an open stock repurchase program provides the company with an additional tool to possibly increase shareholder value. The authorization reflects the board and management's confidence in MARA's long-term strategy and the significant opportunities that lay ahead. Our direct spending will remain focused on site control, pre-development work, and selected project equity where the expected return warrants it. The Starwood model reduces the amount of corporate equity required for qualifying developments, while our liquidity gives us flexibility to pursue opportunities without compromising the balance sheet.
Our focus is on converting power and sites into tenant leases, bringing projects online, and making the earnings potential of this platform visible in our results. Bitcoin built the foundation. Our power portfolio broadened the opportunity. Our partnerships expand the ways we can capture value from each megawatt. Our responsibility now is to execute with discipline and demonstrate the returns these assets can produce. With that, I'll turn the call over to Salman to walk through the financial results.
Salman?
Thank you, Fred. Good afternoon, everyone. Three significant activities define Q3 for MARA. First, after quarter end, we received FERC approval for our Long Ridge acquisition, which we expect to close in the next few weeks. Long Ridge is one of the region's most efficient combined cycle natural gas plants, currently operating at 485 MW, and is expected to reach 505 MW in the first half of 2027. Approximately 70% of its output is hedged under long-term contracts, which provides MARA a unique and diversified EBITDA value proposition compared to our peers. Historically, Long Ridge has recognized over $140 million in annualized adjusted EBITDA, as previously disclosed. Second, we continue to strengthen our balance sheet, deleveraging by reducing $1.4 billion, or 37% of total debt since the beginning of the year.
During the quarter, we borrowed $750 million, collateralized with Bitcoin to fund the Long Ridge acquisition, and subsequent to quarter end, fully repaid the loans, taking advantage of Bitcoin's significant price increase. In August, we borrowed against Bitcoin while it was trading around $64,000, and after quarter end, sold 13,597 Bitcoin at an average price of approximately $81,039, generating more than $1.1 billion in proceeds. Discipline and patient execution enabled us to capture an estimated net benefit of $170 million. Third, we advanced our power portfolio in Matagorda. We hold the rights to develop a site of more than 1,200 acres that ERCOT classified as study load in the Batch Zero process. We expect the site to provide up to 2 GW of grid-connected capacity. We also strengthened the deal itself.
During the quarter, we amended the purchase agreement and revised certain milestones to reduce our near-term cash commitment. As a reminder, MARA did not make payments upfront to enter into this transaction, and we will not incur additional costs to the seller until milestone progress is made. Since quarter end, our board has authorized a discretionary share repurchase program, up to $500 million of our common stock. In the current environment, we are always looking to deploy our capital where it can earn the highest and best return, and at times, that may mean repurchasing our shares. While markets have declined, Bitcoin has remained resilient, and our balance sheet gives us the flexibility to act on those opportunities when they arise.
We have spent years building a differentiated portfolio spanning power, land, and compute, and we remain focused on turning that foundation into contracted returns and durable free cash flow, as Fred mentioned earlier. We will remain disciplined in how we allocate capital, deploying it where we believe it can create the greatest value for our shareholders. During the quarter, we continued to streamline operations and eliminate redundant functions, aligning our cost base with the business we are building. This is the product of patient, deliberate work by our teams, and I want to thank the MARA team for that. With that context, let me turn to Q3 financial results, followed by liquidity and capital allocation. Average Bitcoin prices stayed below last year's levels, though they strengthened over the course of the quarter.
Despite the price environment, we reduced debt, advanced the acquisition of Long Ridge and Matagorda, and kept our cost base in check, positioning us well for the years ahead as a digital infrastructure company. Revenue for the third quarter of 2026 totaled $164.2 million, versus $252.4 million in the prior year period, a decrease of $88.2 million. Bitcoin mining revenue accounted for $83.1 million of that decrease. Average Bitcoin price decreased 37%, lowering revenue by $94.8 million, partially offset by an $11.7 million benefit from higher production. We mined 2,247 Bitcoin this quarter, or 24.4 per day, approximately 103 more than the prior year period. We won 644 blocks, up 2% year-over-year, and down 7% quarter-over-quarter. Year- to- date, we have mined 6,916 Bitcoin.
We delivered energized hashrate of 71.2 EH/s , an increase of 18% from 60.4 EH/s in Q3 of 2025, driven primarily by improvements in miner efficiency. Our share of available mining rewards was 5.4%, up from 5% in Q3 of 2025. Net income was $632.1 million, or $1.28 per diluted share, compared to $123.1 million, or $0.27 per diluted share in the prior year period. This includes an unrealized non-cash fair value gain on digital assets of approximately $885.3 million, reflecting the improvement in Bitcoin price during the quarter. Adjusted EBITDA for the quarter, which includes that fair value gain, increased to $857.7 million, up 117% from $395.6 million in the prior year period. Our cost per petahash per day improved 2% year-over-year to $30.8 from $31.3 in Q3 of 2025. Over the past eight quarters, it has improved by 20%.
We believe that remains among the lowest at scale in our sector. Our cost per kilowatt hour remained $0.04 for our own sites in Q3 of 2026. Purchase energy costs for Bitcoin at those sites was $39,674, up from $39,235 in Q3 of 2025, while production increased 10% over the same period. Third-party hosting and other energy costs, which primarily consists of co-location services at third-party hosted sites, were $71.5 million compared to $75.7 million in Q3 of 2025. The $4.2 million decrease was primarily driven by lower hosting costs, partially offset by higher energy consumption at certain hosted sites. Looking ahead, our most significant third-party hosting arrangements, which represent approximately 29% of our total energy capacity, begin to expire in the third quarter of 2027 and are expected to fully conclude by the first quarter of 2028.
As they expire, we expect to eliminate third-party hosting costs and improve our cost per kilowatt hour over time. General and administrative expenses were $75.8 million, down from $85.3 million a year ago. The 11% decrease primarily reflects a $9.8 million reduction in stock-based compensation expense, partially offset by operating costs of our recently acquired majority-owned subsidiary, Exaion, and administrative costs supporting our expanded footprint. Turning to the balance sheet and liquidity, we ended the quarter with $946.5 million in cash and cash equivalents, and approximately $3.8 billion in combined cash and Bitcoin. Our Bitcoin holdings totaled 34,012 at quarter end, valued at approximately $2.8 billion at a spot price of $83,556. Of that total, approximately 57% or 19,260 Bitcoin was loaned or pledged as collateral.
Since quarter end, our loan repayment has significantly reduced our total loaned or pledged Bitcoin to less than 5% of our BTC holdings. The combination of our cash and Bitcoin holdings gives us the liquidity to meet our obligations, including the convertible note puts, while we keep investing in our operations. Power remains the foundation of our strategy. We have assembled a 1 GW portfolio of owned and operated sites with a clear path to over 4 GW. We have secured the foundation of MARA's digital infrastructure transformation. With power, partners, and capital in place, we are executing our strategy to deliver long-term shareholder value. With that, I'll turn it back over to Fred.
Thank you, Salman. As Salman discussed, our financial results and capital actions reinforce the approach we have described today. We are balancing liquidity, leverage, and dilution while directing capital towards uses that can create the greatest value from our infrastructure. The opportunity is bigger than simply monetizing a megawatt. It is about increasing the value we capture from that megawatt as we move from power to infrastructure to compute, and where it makes sense, integrated cloud services. With that, I will turn the call over to the operator to open it up for questions.
Thank you. We will now be conducting a question-and-answer session. If you would like to ask a question, please press star one on your telephone keypad. A confirmation tone will indicate that your line is in the question queue. You may press star two if you would like to remove your question from the queue. For participants using speaker equipment, it may be necessary to pick up your handset before pressing the star keys. One moment, please, while we poll for questions. Our first question comes from Paul Golding with Macquarie Asset Management. Please go ahead.
Thanks so much for taking the question. It is Macquarie Capital, but appreciate the opportunity to ask around all the progress, and congrats on the progress made so far. I wanted to ask on Long Ridge initially, as we think about the expansion opportunity there, if you could give some color around the maybe emissions permitting backdrop or any other regulatory considerations that should factor in as we think about the recent election, and then I will have a follow-up. Thank you so much.
I think as we look at that particular project, obviously it is already emitting, and if anything, the utilities want us to continue to deliver as much power as we can to the grid, which is something we are committed to. As you know, the majority of the power output of the plant is committed by long-term contractual obligations, and it is one of the reasons we actually found this plant so attractive. Its low cost of operations and the committed revenues make it a great cash-flowing asset for us, and the land was very complementary to the Hannibal site that we already have. So, we have 200 MW via grid connection at Hannibal, the ability to put 200 MW behind-the-meter, which is something that, from a permitting perspective, the team has been working on for quite some time already, anticipating the closing of the deal.
And then additional capacity through both expansions of plant and additional grid capacity. So, that site could reach a total potential of somewhere around 1 GW. We do not foresee any real regulatory issues there. It is a former aluminum smelter plant site where the data center will be built, and we believe that, based on everything we know today, there should not be any real permitting delays for the site.
Thanks so much, Fred. If I could pivot to Exaion, congrats on the new GPU-as-a-service deal. I wanted to ask around some of the back of the envelope on the $1 billion of demand lining up with the 30 MW, I think, that was in the press release shareholder letter. If we do some back-of-the-envelope math there, it kind of pencils out to a pretty competitive mid-30s millions per megawatt of ARR. It is just a quick back-of-the-envelope math. Is the pricing environment that you are describing really pushing through in the form of GPU-as-a-service rental rates per GPU hour? Is this a function of shorter-dated contracts or longer-dated contracts? How should we think about the discussions that are underway and how the pricing strength factors into that? Thank you so much.
Sure. Yeah. So, obviously, the shortage of capacity in the U.S. has forced many prospective customers to look elsewhere for power. One of the benefits of operating in France, and especially with Exaion, is its close relationships to EDF, and France is a country with lots of power availability. Exaion already has data centers with capacity. This specific contract, the initial one, is quite small compared to the other demand that we have been receiving. The contract term, being a little over four years, is fairly average, I think, in the market right now for this type of service. And the pricing is quite good. If you look at the demand curves, there are obviously some customers who want to run Vera Rubin. There are some customers who are more than happy running prior generation GPUs.
And we believe that in the current environment, if you look at the pricing dynamics, the delivery timeframes, the risks of being an early mover with Vera Rubin, using, for example, as in this contract's case, B300s makes all the sense in the world from a financial term because the margins are really attractive when you do it with that type of technology. Exaion has this additional benefit that they have been operating for years a kind of deprecated infrastructure architecture, where they can use GPUs for model building and then deprecate them to inference and then deprecate them to rendering. So, one of Exaion's products and platforms is something called Exaion Studio, which is a rendering platform using GPUs that companies doing advanced media solutions use for rendering 3D graphics, for example, for things.
As an example, you look at the graphics that you would see at the Sphere, for example, in Las Vegas, those types of graphics. Real-time rendering for those types of events, as well as rendering that would be used in films and other applications like that. You can take a GPU and dramatically extend its life across a series of customers. Exaion has built is essentially an orchestration platform that lets you do just GPU rental or more involved orchestration across model selection, inference as a service, all the way up to full private cloud, as well as these type of rendering services we've talked about. It's a great way to extend the life of infrastructure, and it allows us to move all the way from powered land, if you would, all the way to the private cloud.
Got it. The pricing is indicative of how far up the stack Exaion is going as well.
Exactly. Yeah.
Thanks so much.
Yep.
Our next question comes from Michael Donovan with Compass Point. Please go ahead.
Hi, guys. Thanks for taking my question. Fred, with Kearney Advanced in pre-development and the conditional use permit in hand, what are the remaining steps to signing a lease there? Are the open items primarily related to power, permitting, or project economics?
To not kind of go into the MNPI world, let's just say that the issues have nothing to do with anything really on our side. It's simply some things on the customer side that just need to be dotted and T's crossed. As soon as they finalize that, then that lease will move forward. Generally speaking, that's what we're seeing across the portfolio, is there has been a rush to get things done, and then at the last minute, there are always these conditions where the delay is incurred on the customer side, typically, or the tenant side.
Since the election, we've seen a kind of surge, if you would, of demand, where tenants that would normally not have looked, as I mentioned in my comments, at some of our peripheral sites are now saying, "Listen, if you have power availability, just let me come in." I think we're also seeing prospective tenants being very open to solutions that are a little bit more innovative, that will allow them to deliver faster. I think the other thing that we're really seeing is, and this is feedback we're getting from hyperscaler, for example, is they are spending an inordinate amount of time dealing with delays caused by first-time developers. This is one of the reasons why we believe the Starwood partnership is so important, and is so critical.
Especially hyperscalers who have worked with Starwood before, they know what they are getting when they work with Starwood and through Starwood, us. Whereas when they work with one of our peers potentially, whose first time at doing this, there is not certainty. While many of them may have been attracted by promised timelines, they are now realizing that the promise does not necessarily resolve itself in delivery and in service on schedule. A lot of time is being consumed at the tenant side with dealing with deals that have now gotten kind of delayed, or are stuck due to development constraints and other things.
I am very optimistic about the fact that by having waited and not jumped in from the very beginning, some of you may have recalled some of our early comments were very much about, if we go in and build the types of data centers that people were talking about 18 months, two years ago, we said, "Hyperscaler will not work with a Bitcoin miner." I think hyperscalers have learned their lessons or are learning them now, and we are starting to see a different quality of conversation happening. We think having been patient will pay off in the long run for us.
Understood. You have talked about applying your modular deployment experience to AI infrastructure. How do you think about the design and development? Would MARA and Starwood develop a standardized solution, or would you expect to work with third-party providers using designs such as NVIDIA's DSX pod?
Yeah. I mean, it really depends on the tenant, right? We are not in the build it and they will come business. We are in the business of signing a lease with a tenant who has a very specific requirement as to what they want and who is willing to put their money, if you would, behind them on the technology that they want to deploy. If they want to deploy NVIDIA's modular solution or any of the other various solutions being touted by people, we are more than willing to deploy that for them, but it is their risk if they are the ones driving the decision.
We think there are a lot of very attractive solutions, but I think the key is not just in the design of the modular solution, but in how do you build a site for it and how do you operate a site for it? We operate 250 MW of containerized solutions in the U.A.E. that are based on liquid immersion, that are, when they were built in 2023, state-of-the-art and still operate amazingly. We have a very large, I should almost say one of the largest containerized sites in the U.S., in Texas, and we have learned a lot over the years in how to do this. It's the operations of these sites that is just as important as the technology behind the container, the racking, the cooling, et cetera.
Great. Thank you, Fred. Best of luck.
Thank you.
Our next question comes from Chris Brendler with Rosenblatt. Please go ahead.
Hey, thanks for taking my questions. I apologize, I missed this. I'm doing a little multitasking, but have you updated this on Matagorda and what the election results potentially mean there? I think there's like a, you know, sort of like there's incentives in the contract where you don't have to pay anything until they get approval, but how does the change in government there affect that outlook on Matagorda? Thank you.
Yeah. Hey, Chris. Yeah, it doesn't change anything in the contract. We did this deal very specifically to mitigate risk, and it has been proven that the way we did this was the right way to do it. So, at this point, we have no real obligations until Batch Zero is fully approved, and that's when the obligations start. If you look at between Governor Abbott's audit requirements and kind of the environment today, more than likely, Batch Zero approval or studied load will get you power at about the same time, if you look at the time frames. As you know, the new governor of Texas has said that projects that have been approved are going to move forward.
We believe that Texas as a state will continue to be very cooperative with the industry, though they will want to ensure that consumers aren't being harmed, et cetera. We look forward to working together with communities in Texas to move this forward. But Matagorda is an amazing site. It's very close to a nuclear power plant. It already has all the high-voltage infrastructure coming into the site. It's in a great market for power pricing, even to the extent that we can mine Bitcoin there profitably. So, it is a great site for us. We're very excited, and we believe that as we go through this process, it'll become a great campus for us.
That's very helpful. Thank you. I guess my follow-up is actually on Bitcoin mining business.
You are still growing your hash rate, then there is some weather and the summer heat lower the effective hash rate—
Yeah.
—in the third quarter.
But it did come in, I guess, lower than I was expecting. Are you, at this point, I guess you are optimizing still on the fleet. How are you thinking about optimization once the hosting sites roll off next year? Is a lot of that work already done where the older machines are already at some of the hosting sites, or you have some more work to do on moving the more efficient fleet into your existing sites? Thanks.
Yeah, great question. We have been slowly upgrading older systems. If you think about the age of our fleet, we started deploying miners at scale back in really 2022, and so these older machines are coming up on the kind of their end of useful life in some cases. And so we have been upgrading to modern, more efficient machines, especially taking advantage of the extremely low pricing of Bitcoin mining hardware today. And so that is why you have seen hash rate climb up, just because more energy-efficient machines produce more hash rate per megawatt, right? We have been doing that. As you look at the hosting deals rolling off beginning next year, a lot of that capacity will be replaced through machine upgrades at other sites, which will maintain the hash rate, if you would.
And then as new sites are being prepared for development into hyperscaler sites or AI campuses in general, where we do intend to deploy modular technology, we can easily deploy Bitcoin mining in modular format and literally move stuff from one site to another. That's part of the whole idea of the benefits to doing modular, is the Think of it as the pad infrastructure is very similar between the two. While the containers may be different, while within the containers may be different, everything but that is quite similar. That's an area where we think we have this ability to have very flexible use of sites.
Makes sense. Thanks so much. Appreciate it.
Our next question comes from Ben Sommers with BTIG. Please go ahead.
Hey, good afternoon, and thank you for taking my questions. First, on Exaion, just kind of curious, you mentioned how you plan to scale the power with your growth goals there, but can you just talk a bit about GPU allocation and kind of how you think about purchasing GPUs to kind of help meet those growth goals?
I think here is kind of a way to look at it. You have customers who have a lot of demand for GPU capacity. You have data center capacity not growing at that same rate because of just practical limitations of getting these sites permitted, energized, and on. You have GPUs that are sitting in limbo that have been delivered but are not going to be installed because the sites are not ready. You are starting to see some inventory builds where people who have procured GPUs are now thinking maybe what they should do is sell what they have and then reorder next-gen equipment possibly. We are not finding any limitations in the GPU market today. We have spent time building relationships with NVIDIA and a lot of their partners.
I was just recently up at NVIDIA headquarters, and we spent time, the Exaion team spent time with the executive team at NVIDIA at GTC in Berlin. We are very much in the relationship-building mode with them. We have a great relationship with them, and so far, we have not seen any issues with the ability to procure the volume that we need.
Super helpful. I know we have some access today to some power internationally and could potentially expand. Is there any thoughts of developing properties with Starwood and then having Exaion potentially plug in GPUs and kind of having that fully vertically integrated stack?
Yeah, I mean, that is exactly what the strategy is, right? Think of us as a developer of a strip mall, right? You have an anchor tenant that is a hyperscaler, then maybe you have some smaller tenants that are enterprise customers. Exaion can operate that infrastructure on their behalf if they need it. Then in between, you may have tenants that are just providing AI services. The whole model is kind of designed around being able to start with land and power in the base and go all the way up to private cloud and operating enterprise infrastructure. Or we have the flexibility with Exaion that they can go in and operate an enterprise's infrastructure or essentially provide inference as a service and other services to enterprises and other providers as needed. One thing I just did want to kind of double-click on.
Hosting capacity that Exaion has contracted with the U.S. customer is based in France. Meaning, the U.S. customer is in the U.S., the hosting is happening in France. That is an example of the type of demand that exists out there for people. If you have capacity and you can turn it on today, there are a lot of customers waiting in line to take that capacity.
Super helpful. Thank you for taking my questions.
Yep.
Our next question comes from Mike Colonnese with H.C. Wainwright. Please go ahead.
Hi, good afternoon, guys, and thank you for taking my questions. Great to see all the progress here. First one for me, maybe Fred. Can you discuss how lease terms under negotiations have evolved given this change we have seen in the broader macro backdrop in recent months? Has the calculus on potential contracts changed at all with regard to the ownership position MARA is willing to take in a potential deal with Starwood?
I think the way to look at it is, most of the tenants no longer are relying on promises, but are double-clicking and triple-clicking and doing their diligence deeply to ensure that a prospective developer can actually deliver on time. So, digging in deeply on power contracts, digging in deeply on environmental permitting risk, et cetera. Part of the way we get ahead of that is we move forward on those things even though we don't have a signed lease. We believe enough in our ability to execute that whether that tenant signs or somebody else does, all of the work that we do, the pre-development work, will benefit from a timing perspective, meaning the site will come online even faster, right? There's no reason to wait.
We're not going to start building a building without a tenant, but we'll do power infrastructure, all of that stuff, network, et cetera, so that we're ready to go with that. So customers, the tenants are being much more demanding in their diligence process, which we think is great. We've had lots of visits to the sites with prospective tenant operations teams, engineering team, network teams, et cetera, which means they are investing their time and resources to try and validate those sites are good, and we're getting great feedback from that. As I mentioned earlier, some of the feedback we're hearing, especially from the hyperscalers, is having signed deals with first-time developers, they're now having to spend a lot of time fixing those deals and pushing those things forward. So, as with the Kearney site, for example, it's public regarding the permitting there.
We've worked very closely with the community there. We have a great relationship with the community. Things are moving along greatly there, and we believe that site will be amongst the first ones that gets contracted.
Thanks for the color, Fred. Salman, maybe one for you. How should investors think about the pace of Bitcoin sales from here as you look to fund AI data center developments and potential share repurchases? It looks like you have just over 35,000 coin on the balance sheet right now. I know it's been very topical with the investment community around your exposure to Bitcoin price action. Just any color there would be helpful.
Yeah. We've always said that Bitcoin is a strategic asset and corporate asset on our balance sheet that will be used for investing in our company and as a source of capital. What you've seen play out since the quarter ended is that we have sold Bitcoin and paid down our borrowing against Bitcoin and delevered the balance sheet about 37% since the beginning of the year.
Obviously, strengthening our balance sheet further overall from a net liquidity standpoint and also build a cash or a capital source through the Bitcoin that we already have that can be utilized for investing in the highest or the best use of that capital allocation, if you like, at that point in time, which includes paying down debt, or as I said in the prepared remarks, the puts, if they are put to us, we have enough cash and liquidity to be able to entertain that. On top of that, we have amazing projects, as Fred talked about a few moments ago, in diversified communities and diversified states, multiple countries. That gives us ample amount of leeway or bandwidth to continue to invest in high-return assets.
Just as a reminder, because of our partnership approach, which is unique to MARA versus our other peers, the strategy for us is capital light. If you do the math, we can go multi-gigawatt build opportunity through the existing liquidity that we have on the balance sheet through the Bitcoin that we hold and the cash that we hold as of now. That puts us in a very unique position without having to think about the financing, the traditional financing of these transactions. The net check that MARA has to write within a partnership scenario is a capital-light strategy, which gives us a long ways to go from here from a development standpoint.
All in all, look, Bitcoin is a strategic asset for us, and we're utilizing it methodically, and that gives us the opportunity to, as we also said in the prepared remarks about share repurchase. That is another option or a tool. Should the right opportunity arise, we will utilize that appropriately and responsibly as the market conditions dictate.
Got it. Thank you, and best of luck with advancing the discussions.
Thank you.
Thank you.
Our next question comes from Kevin Dede with JonesTrading. Please go ahead.
Thanks for having me on the call, Fred and Salman.
Hey.
Curious about the Exaion deal.
We do not have an RFS date, and it is not clear. I know you talked about all sorts of different classes of GPUs.
That is not clear. The power is not clear. The potential scale to go to $1 billion or $3 billion in contract value, I am just kind of wondering how all those pieces come together when it comes time to allocate power resources.
Yeah. So, great question. I think it's in the public domain that the GPUs that are powering this initial contract are B300s. Those have been ordered, and delivery will be received commencing this month through the first part of December, and we expect the RFS date on the first part of this contract to happen this side of the new year. So, we'll have a portion of this capacity up and running by the end of the year. So, that gives you an idea of the speed at which we're able to move, the ease with which we're able to acquire GPUs as well. This goes into capacity that Exaion has access to in France that's already built, it's already ready to go. Again, part of the benefit of this strategy with Exaion is, in the U.S., there's huge scarcity around capacity.
In Europe, if you do things in the appropriate way and work with the appropriate partners, there is ability to get access to capacity. There is ability to get access to power. There are some short routes to do things. Also remember that Exaion operates in Canada, and we have similar opportunities in Canada regarding power and infrastructure. So, the Exaion build plan, if you would, is not predicated on MARA developing anything in the U.S. for them. This is really predicated on them building capacity on their own in Europe and in Canada, which they are very experienced at doing.
Okay, that's helpful. My understanding was that, I think there's a site in Sherbrooke, but it's de minimis in megawatts, and I just know Hydro-Québec is really pretty tight-fisted, and I'm kind of wondering if you could add some color to that.
I'm not going to criticize Hydro-Québec. Exaion has capacity available both in Canada and in France. There's actually, France has some nice availability. The other advantage in dealing with France is that they have excellent internet infrastructure. You have a hugely reliable grid system. You have over 60% of the power is clean nuclear. You have a lot of renewable energy. So, essentially, for companies where carbon footprint, which I think will become a very important thing again here over the next coming years after the election results, has all of a sudden become something very important, and people want to ensure that they're able to operate AI infrastructure in a way that is clean, that doesn't generate unnecessary carbon emissions. So, that's also driving some of this.
France has a lot of energy that is not being used today, and infrastructure has been built that has not been fully utilized. We are finding lots of pockets where we can jump in and take capacity, which is great.
Okay. Thank you. Thank you for the perspective, Fred. Appreciate it. Hard to see plants from here.
We have reached the end of our question-and-answer session. I would now like to turn the floor back over to Robert Samuels for closing comments.
Thanks, operator, and thank you, everyone, for joining us today. If you do have any questions that were not answered during today's call, please feel free to contact our investor relations team at ir.MARA.com. Thanks very much and enjoy the rest of the day.
This concludes today's teleconference. You may disconnect your lines at this time. Thank you for your participation.