Bitcoin Investing in the Age of AI: Why Miners are Pivoting w/ MARA CEO Fred Thiel

An interesting phrase like mullet data centers where AI in the front, then you have the Bitcoin party in the back. That's, you know, shutting on and off the AI business. You have power in the bottom, then you have chips, then you have infrastructure, then you have system software, then you have application. We believe that owning that bottom layer or owning the control of that bottom layer is critical. All of the same stuff we did in the Bitcoin world. Oh, I've got a megawatt of power. You know, I'm running machines at 50 joules per terahash. Well, if I have a machine that runs at 25 joules per terahash, I now just doubled my Bitcoin mining capacity on that same amount of power. It's the exact same dynamic that is happening in the AI industry. Best thing for us is we get to operate those sites, Bitcoin mining sites, until the AI comes online.

Great to be here.

Yeah, I think if you go back to our Q3 earnings call last year, which was I think in September, we felt the market was a little frothy. You know, between the digital asset treasury companies and the ETFs, there was a way overblown kind of demand in the market, which wasn't normal. And if you look back to the early days of Bitcoin to today and you normalize price on that curve, you'll see that we've kind of reverted to mean and we're still adjusting back. The price area of the mid-50s is kind of where there's very strong support for the price of Bitcoin at this point. And what's going to drive further price appreciation is, you know, very macro. Bitcoin is highly correlated to macro and because it's a risk-on asset in most people's minds, it's correlated to what's happening in the stock market. So if Trump says we're going to own the Straits of Hormuz, the market goes down, Bitcoin goes down, the dollar goes up. If we all of a sudden have peace with Iran and inflation is tame, then you know dollar goes down, Bitcoin goes up, and stock market goes up. So this kind of you've got this macro market correlation and you've got this risk correlation and equity market correlation and unfortunately the geopolitical world and the political world right now are causing a lot of yo-yos. And you know, Bitcoin is a market where because Bitcoin doesn't provide any native yield, it is a place where people want to put their capital. And, you know, if Bitcoin starts assuming a number of days of continual appreciation in price then it builds momentum, and you know we saw that for a period of time earlier this year where it started clawing back and was, you know, we came from the 50s back up into 60s and we're progressing well up there. But I think right now it's very driven by exogenous events and I don't think that if clarity bill were to pass, I don't think that's going to have a huge impact on Bitcoin. I think Michael Saylor has provided a little bit of an overhang with stretch and that whole model. You know, when you've got somebody who holds nearly 900,000 Bitcoin on their balance sheet, or maybe over 900,000 at this point, the fear of a liquidation event is, you know, definitely going to fuel a lot of shorts. And so, you've just got a lot of these events going on with Bitcoin. And I think unfortunately we're in this situation where because there isn't an actual beneficial business use case for Bitcoin other than as a store of value, you have to wait for people to believe it's the right place to put their money again. And right now they've been chasing AI stocks that have been going up and they've been chasing Starling, I mean SpaceX, and things like that. You know, long-winded answer, but I think you've got a lot of exogenous events driving people's perception of the value of Bitcoin. And we need people outside of the Bitcoin community to view Bitcoin as valuable for Bitcoin price to go up.

Yeah, I don't think he has any specific policy related thoughts on it other than if there's any risk to undermining the dollar, which I don't think he does. I don't think he views there's any systemic risk in Bitcoin necessarily. So, I think that's, you know, he's very focused on dealing with inflation that is caused by macro events caused by the government. At the same time, you know, unemployment isn't in a bad spot, right? So he's kind of got to deal with inflation, but he doesn't want to necessarily cause unemployment numbers to go up. And then you've got AI throwing its spectre on those two things that the government measures. And you got to realize that, you know, companies are spending a lot of money on tokens right now and not crypto tokens, but, you know, Anthropic and OpenAI and Google and others, and that is impacting inflation, right? So you've got a lot going on that he's got to deal with. And it's a very unstable world right now. Geopolitically, governmentally, got midterm elections coming up. The Fed has to be seen as not being politically driven. And I think by changing his bias regarding how the Fed is going to communicate, I think it's going to give him a little more freedom to actually do what he's got to do as opposed to what some of his predecessors have had. So, I think we'll see. It's a time where I think all business leaders and investors need to be very focused on being agile and being resilient. And you know the benefit that Bitcoin has over gold is the fact that it is very easy to move in and out of Bitcoin. It's a lot harder to move in and out of gold. You can't physically easily take your gold with you. You can take your Bitcoin with you wherever you want to go. And I think over time the narrative around digital gold will take hold. I just wish that there were more commercial use cases for Bitcoin that would drive people to transact more with Bitcoin. Because the greater fear for the Bitcoin mining industry is that, you know, another couple of halvings, power prices aren't going down, and if Bitcoin price doesn't go up commensurate with the halvings and you don't have transaction fee increases, then it will become hard for miners to support the level of mining that's being done there.

Well, I think as you look at AI, there's obviously a very large capex spend, you know, 600 plus billion dollars this year and an estimated, depending on the analysts you listen to, potentially a trillion dollars plus next year. That obviously has an impact on the economy. It has an impact on construction jobs, on copper prices, has impact across the board. It has an impact on the stock market, which drives people's feelings of their level of affluence, right? If your portfolio is going up, you feel rich. If your portfolio is going down, you feel poor. And if you look at the Fed, inflation, there's not a whole lot they can really do to counter inflation. Interest rates alone are a less than perfect metric these days because so much of the federal deficit is not tied to commercial markets at all. And then because of the high interest payments the government has, you have fiscal, real fiscal spending issues that are not in the areas where the government can just change things, right? It's near impossible to change the entitlement structure in this country, like in any country in the world most probably. So I think what you've got to worry about is you have inflation which is being driven by these exogenous events. If you lower interest rates and try to speed up the economy, does that mean that people are going to invest in more factories and create more jobs, or what is the world going to invest in? And right now the single biggest area of investment in the corporate world is around technology and AI, right? You have all this technology going on in the space. You have technology going on in AI. You have technology going on in communications, which is a subset of the space race because of the satellite business. And that's going to continue. The AI industry will have as big an impact, if not bigger, than the internet did or that a lot of other technologies have had. And it's very infrastructure heavy, where the internet wasn't necessarily as infrastructure heavy. AI certainly is. And it takes years to realize the benefits of that infrastructure. And I think companies are going to become more productive in the near term as opposed to more efficient, and I don't think you're going to see massive job losses due to AI anywhere in the near term. But I think over time you'll see a redistribution of employment where people who are doing certain types of jobs will migrate to other areas, not unlike transitions that have happened before, because in this case you're not offshoring jobs, right? The jobs are still going to be here. They're just going to be different.

So today you have less power available than the data center operators would like to have for the compute that they want to bring online. So demand for AI drives capacity needs. Capacity needs drive needs for power, and the capacity needs and demand for AI are outstripping the available amount of data center capacity there is today. And it takes a couple years to build a data center. More importantly, the data center has to be plugged into power, and power plants take 6 to 8 years to build. So there's a floor to the cost of power too, right? So if you take power away from consumers to give it to the data center world, then the pricing of power to consumers is going to go up. So you obviously have a lot of dynamics going on today driving decisions that are being made in the power industry. We believe that if you think of Jensen, you know, the CEO of Nvidia, if you think about his pyramid of hierarchy of the AI business, you have power in the bottom, then you have chips, then you have infrastructure, then you have system software, then you have application on top of that. I'm generalizing here. We believe that owning that bottom layer or owning the control of that bottom layer is critical. And then we don't think owning chips is critical. So don't see us saying we're going to be a neocloud anytime soon. And then you have above that which is infrastructure. So think of that as the actual data center, the cooling, all that stuff. So you have power, land, you have infrastructure, you sandwich chips into that and you have a data center essentially. Very simple. Power is a very important thing to have because you will always need it and the price of power isn't going to go down. Whereas the price of compute and the cost of providing that compute to the marketplace will be under great pressure going forward, I think. The silicon world is seeing a lot of competition, but because there's so little power today, if you have silicon as old plugged in, you're still getting reasonable value for that silicon. But imagine a day where you have a lot more power available and all of a sudden that silicon, the competitive aspects of that market are going to impact it. When you have a restricted market like you have today, meaning a constrained market relative to capacity, what does that do? It drives people to look for more efficient silicon per megawatt. It drives people to develop software stacks that are more efficient. All of the same stuff that we did in the Bitcoin world, right? Oh, I've got a megawatt of power. I'm running machines at 50 joules per terahash. Well, if I have a machine that runs at 25 joules per terahash, I now just doubled my Bitcoin mining capacity on that same amount of power. It's the exact same dynamic that is happening in the AI industry, only it's not yet really being felt in quite the same way as it will be once more power comes online. And so our focus is very much to acquire access to significant amounts of power. We've this year more than doubled our total power under control when we close our Long Ridge Energy deal later this year. Together with the deal that we just announced recently, we'll be well over 4 gigawatts of power in our portfolio, contracted and owned power. So we feel very comfortable in that spot and think that that is a great place for us to be. We then partnered with Starwood. Bitcoin miners are really good, and this is something I've said for a long time, Bitcoin miners are great at building tier one data centers, but we're not good at building tier 4 data centers. And so, if you think of it kind of this way: if you have a loved one who's sick and you're going to take them to a doctor, you're going to take them to a doctor that's done that particular procedure thousands of times over or somebody who's never done it before. It's the same in the data center world, right? If you can partner with somebody who has built data centers for hyperscalers, who has done it on time, to spec, and has been a great counterparty, then it's a lot easier to get deals done. And so we decided that it was better for us to partner with somebody who had built over 7 gigawatts of data center capacity for hyperscalers, who has captive construction, and has a track record of doing so successfully, and that's why the relationship with Starwood is so ideal for us. It's also very capex light. Unlike our peers who have to put up all of the money to do these projects, we contribute a site into a venture with Starwood, they have to catch up to our contribution before we have to put the next dollar in. And so it lets us leverage the combined benefits of both of our best skills. And the best thing for us is we get to operate those sites as Bitcoin mining sites until the AI comes online.

Yeah, I think there are lots of interesting aspects of blending the two. For one thing, you have just the time shift benefit of, we can populate a large site with Bitcoin mining in a very short period of time while the actual data center for AI is being designed and built. So we can have something live in months versus years. You figure it takes 18 months to 24 months to build an AI data center; we can have a large site live in 12 months in the Bitcoin mining world. The best thing about that is that the Bitcoin mining is all containerized, so as the AI data center comes online, you just pick up containers and move them to the next site. So that's a great use there. When it comes to the flexible load aspect, the power industry has only about a 2% window in the day where over 98% of power is actually used. The rest of the day averages about a 60% load, which means there's a ton of capacity available in the grid today in the US that isn't being used. The problem is it has to be able to curtail 2% to 3% of the time, and Duke University did a study about this last year which is great because it truly identifies this issue. So it's not about having to build lots of new capacity; it's about having to operate capacity in a way such that you can operate it when there is power and then curtail when there's need for power on the grid. Part of the challenge is that hyperscalers aren't used to operating in that world, and they would prefer not to have to. And because the margins are such, the AI industry, the cost of power is but a small piece of the cost of the actual data center; the infrastructure and the stuff that goes in the building is a much bigger piece of that. So when you look at the opex, the power is still somewhere less than 10%. When power gets up to be 50% to 60%, then there's a motivation to start looking at how do we get curtailment payments, how do we get flexible load status, how do we do this, how do we do that. And then it's a question of the world of Bitcoin mining still has to be an attractive world to use the extra power and then be able to curtail it. Right? So if you think about Bitcoin economics, unless you're using machines that have been depreciated, running only 30% or 40% of the day is pretty tough to make it potentially viable.

Right. So it's you have to look at the economics of the alternative use for the energy at the same time. Right? So I think there's a lot that has yet to be kind of discovered in optimizing AI data center workloads, how they're orchestrated, how power is orchestrated around them, and how do you do proper load balancing with them. We've developed some very interesting technology that we call Vertebrae, which load balances our own data centers. It does load following, meaning it's able to run a data center in a perfect follow curve to a power source. So we own a wind farm, we can mine Bitcoin on the wind farm, maximizing operational efficiency because we can ramp our production up and down together with the wind as opposed to having to have just this lower level consistent mining that goes on. So as we bring that technology to market on the AI side, I think it'll be very interesting to see how it can help lower the cost of running inference, which are the ones that are consuming all the tokens, which is where people are having challenges today with payments. Plus, I think you're looking at a world where because using tokens on frontier models are so expensive today, people are really looking at running open source models on infrastructure where they can run all of their analyze these emails, look at this financial statement, all the bread-and-butter stuff you do with AI that you don't need to use frontier model for. Plus those applications where those use cases where you're not going to put your most valuable data up into the cloud because, look at it today if you're using Fable, they retain your data for 30 days. Well, that's not good if it's your proprietary data. So that was one of the reasons why we made the investment in the French company Xion because they operate private cloud. They know how to do everything at the other end of the spectrum from the hyperscalers. And so we believe over time those marketplaces are going to come together.

A lot of what we learned on the Bitcoin mining side is directly applicable to the AI side. Fear comes due to a lack of knowledge, right? You're afraid of something because somebody has told you you should be afraid of it. But when you educate people to the reality, I think you can win over a lot of people to see the benefits. There are communities where we operate where there are not a lot of jobs and not that we have generated thousands of jobs in the community, but what we bring are tax revenues that are very significant. And so our tax revenues support the first responders for these towns. They support lots of activities and if we weren't there, they would have to tax somebody else for it. So it's a lot of educating the populace of these communities that hey,

Listen, we understand that you wish maybe we could do something to attract thousands of manufacturing jobs to a community as opposed to hundreds of data center jobs. But you have to really look at the financial benefit we bring, or what we call the external rate of return of what we do. So there's an internal rate of return, how we're returning value for shareholders, and then there's an external rate of return, how we are generating value for the communities where we operate. I mean, look at the GDP. Studies have been done on this in Texas, for example, third-party studies where the economic impact of our investing in a community is huge and follows on down the street for many, many years. So I think it's an issue of education and being very proactive working with these communities early upfront. I think there are definitely parts of the US where communities have been turned into data center parks and people don't necessarily like that. And that's all a question of where you put these data centers such that they're not impacting communities. A lot of areas that are industrially zoned are perfect for building data centers, but then you have the fear factor of the data center. Are the computers going to cause radiating damage? Are they going to cause other things? Is there noise pollution? Is there water pollution? What's going to happen to our water? What's going to happen to our power prices? And I think a lot of those are questions that just need to be answered so that people can be educated and understand that a well-run data center recycles its own water, so it's closed-loop water. So you have a one-time use of water to cool it. I think you're going to see non-water-based cooling solutions coming to market or being deployed. They're already in market but being deployed over the next few years that will solve a lot of that headache. And I think between efficiency of silicon and efficiency of software, you're going to see the energy efficiency of compute dramatically rise, but that will also drive continued increase in use. And I don't think a lot of people really understand how early we are in the adoption process of AI. Most people don't run agentic frameworks themselves today. Most people aren't using agents to any greater extent. It's still a very narrow segment of the market that is driving all of the demand for that. And over time, just like the internet, just like cell phone penetration, it takes time. People's use of the internet beyond just browsing took time, and the internet went through multiple phases. The AI industry is going to go through the same thing.

Well, I mean, I'll give you a practical example, which is not a new one. It's a number of years old, but when pathology operations in hospitals started using machine learning to analyze slide samples. So pathology works: you go in, you have a biopsy, they take samples of a piece of tissue, they create slides, put under a microscope, the images are taken, and then a radiologist historically would look at the images and say, 'Is this cancerous? Is it not cancerous? What's going on with this tissue?' So a number of years ago, MIT started a machine learning project where they ingested huge amounts of slide samples and then had this AI-assisted pathology system recommend to the pathologist, 'Look at this slide, look at this slide.' And you still had a human looking at it. Then Stanford did a subsequent study where they were able in six weeks to teach the model through reinforcement learning even more ability to detect cancers. And I think even that study discovered two new types of breast cancer that they hadn't seen before.

It's not until we get to the point where we trust AI enough to operate full end-to-end workflows that we're going to be in a place where you really start seeing these huge efficiency gains. But realize that the developed world has a demographic challenge facing it, which is declining birth rates. If we want GDP in the world to continue to go up, we have got to have a way to make humans much more productive. Or else you're going to see GDP decline, tax revenues decline, benefits are going to have to decline, prices will decline, and you get a deflationary spiral. So everybody should be betting on the fact that they really want AI to be successful because it's the only way we're going to drive growth in GDP with a declining population.

Last point on this is that, look at self-driving cars, right? When was the first self-driving car kind of announced and where are we today? A lot has to happen before you can take the human out from behind the driver's seat. There's still lots of things that AI is going to tackle where you're going to have this kind of five-plus-year process of getting to a point where we trust the AI enough to do it. And meanwhile, AI doesn't have this human blocking constraint on learning. It can teach itself at a speed which is only constrained by the amount of compute that it has and the amount of data that it has. So I think you're seeing this reinforced iterative learning that the frontier models are going through now, which is why you get a new one every couple of months as opposed to every six months, and soon it'll be basically they could release a new one every day because it's only limited by the speed of the AI to improve itself. And I'm very optimistic about what AI is going to bring to the world. I'm less optimistic about how successful humans are going to be in enabling that. I think humans are going to be the blobby factor for quite some time.

You have to put it in context of the fiscal world we exist in. We operate in a world where governments operate deficits that are getting larger. They are operating with large debt loads which need to be serviced, and more and more of tax revenues are going to pay those interest payments. All of which says the only way out of that problem is to inflate your way out of it. And the only way to do that is basically print money and do all sorts of things that the government likes to do to drive inflation. All of which speaks to Bitcoin being a form of digital gold, ideally as a place where people put their capital to hold it. Because hard assets become very valuable in inflationary times, right? Land, property, art, things like that -- anything that has scarcity associated to it. In a deflationary environment, you need to watch out because hard assets are worth less because there are fewer people to want things, so there's less demand. So I don't think Bitcoiners are interested in a deflationary world. I think if anything, Bitcoiners should be interested in an inflationary world because that's the best narrative for Bitcoin at least today. But look at gold. Gold spent many years relatively stagnant and then all of a sudden it had this big run-up as people were getting more and more afraid of inflation. Bitcoin ran before gold, as Bitcoin tends to do. It tends to run six months ahead of gold. So I think we should expect greater inflation in the future, or we're going to see fiscal challenges, all of which I think is going to drive people to want to put their money in Bitcoin over time. And so I'm confident longer term about Bitcoin. I think just right now we have some short-term issues that are geopolitical and macro.

The exchanges of value require both sides to agree on the value of what it is they're exchanging. Right? If you go from the barter world where you had eggs and I had milk, we have to decide how much milk I can give you for me to get a dozen eggs. You have two different perspectives on the value of the milk. What money brought was a unit of exchange that everybody agreed to its value. A dollar is a dollar. You know what you can do with a dollar and I know what I can do with a dollar. So we have a mutual understanding on the value of the dollar. All of which goes to you want to use a medium of exchange that is not volatile, because a volatile medium of exchange, if it's volatile, when I pay you and then you go use that to pay, it has changed its value. You could potentially be in a situation like any hyperinflationary environment where you and I exchanged Bitcoin for a good or service at $100,000 and then now it's worth $75. You don't want that. So agentic systems will naturally tend to go towards stablecoins because there isn't a yield case around Bitcoin and there isn't an appreciation case in the buying and selling of goods and services that agents are going to do. That makes sense. And then you have speed of transaction processing speed and transaction fees and all the other things that get into it. So my belief is that agentic systems will tend to use some stablecoin variant to do most of their transactions. That being said, stablecoins may decide to store their value in Bitcoin. I think that's going to be very interesting to see. Is it that we're backing stablecoins with US Treasuries? At what point given the fiscal situation do we end up including Bitcoin in that mix and stabilizing the value of treasuries through something that is volatile but often volatile to the upside? The Treasury reason for stablecoins has everything to do with what the US government wants versus what makes fiscal prudence for a stable issuer. The federal government wants stablecoins to buy as many Treasury bills as possible because other foreign governments are not buying as many.

Um, I would give a little bit of color that has nothing to do with my being involved in Bitcoin. It has everything to do with I was at a point in my career chairman of one of the largest European cryptographic companies. And we were asked at the time by NIST, the National Institute of Standards and Technology, to contribute to a process of defining quantum-proof encryption technologies. So we were involved in that together with IBM, Google, Microsoft, and a handful of other companies over 10 years ago. Four algorithms were recommended; three of them remain as the quantum-proof or post-quantum encryption recommendations. So in that process, you learn some interesting things.

The underlying asymmetric technology that makes up the encryption used for Bitcoin wallets primarily -- because the Bitcoin blockchain itself is not at risk, it's wallets that are at risk. When you look at that, everybody knows that the Bitcoin blockchain is so transparent that if one coin moves in a wallet that's been identified as a lost wallet or a Satoshi wallet, the whole world is going to know about it. What most people don't realize is for a decade plus, certain powers have been aggregating information from internet traffic that they don't have the ability to decrypt because it's all done under SSL. Well, when you get a quantum computer, the first thing you're going to do is not attack a Bitcoin wallet. The first thing I would do as a superpower with a quantum computer is decrypt all of the login that people have done to their bank accounts, to investment accounts, etc., which are all of a sudden plain text to a quantum computer. And I would go ahead and log into accounts, take 10 cents out of every single account. Nobody is going to complain about losing 10 cents. It will take a long time for anybody to figure it out. And I'm going to take a lot more money than moving one bitcoin and all of a sudden having the whole world know about it.

And so the banks, Google, Microsoft, all of these are much more worried about that quantum threat than the Bitcoin quantum threat. The minute you see the security certificates on major websites change to quantum-proof, then you know it's a real problem that's about to happen. And there's a big cost to them doing it because every single login to every single system today generally speaking is asymmetric and is totally hackable by quantum. They don't have to hack your password by brute force with a quantum computer. They just decrypt all of the internet traffic that's been going on and they can do it. So I think part of the problem is that people don't really understand what quantum computers are. They are not general purpose computers. They are good for very specific things, essentially multifactorial analysis: where is the hurricane going to go? Let's look at all the permutations. What are the theoretical results of this type of drug use? Asymmetric encryption is a factoring problem which typically uses brute force to attack. With a quantum computer, you know the answer. That's what it's really good for. Is it good for transaction processing? Absolutely not. And you're starting to see advances on the CPU and software side that are really improving traditional CPU capabilities. But you haven't gotten to a point yet with quantum where you are at a low enough error rate because error correction is really important. Google's latest estimate is 2029, other people say 2030. I think we're still a few years off, but that doesn't mean people should be complacent. If you have a Bitcoin wallet that is an early Bitcoin wallet, you should generally, even if you have a new one, don't leave your Bitcoin sitting in a wallet you transact in and out of because the wallet address is now on the blockchain. Half of your key is essentially exposed for a quantum computer. So whenever you transact, always transact into a new wallet, never leave the balance in the existing wallet. So I think it's about education. Part of what the Mara Foundation is really focused on is educating people around security aspects of Bitcoin, working with core developers to put in place improvement proposals to deal with the issue. But Bitcoin is a community where you've got to get a lot of consensus for anything to pass. So it takes a lot of time to get things to pass. But I don't think the community should do anything abrupt like soft forking the 1.6 million wallets that are out there that are susceptible to quantum, including the Satoshi wallets. I think there should be a lot of hard thought put into that, and that's a lot of what the Mara Foundation does: focusing on educating and working with experts and people to understand what is in the realm of the possible, what is in the realm of the likely, and then recommending things and working with people to put in place the best solutions.

Well, I think the balance of the year you'll see us sign some tenant leases, which is proof of the strategy. I think the market will realize the benefit of the joint venture approach, it driving a lot more certainty in the outcome and making it much more capital efficient for us. Because I do think that you see a lot of projects that have been delayed in the data center world. A lot of that is due to people trying to do it themselves, and I think partnering with somebody who has been there, done that, is a trusted counterparty to hyperscalers is going to play to our benefit. And we remain very interested in not just the domestic but the international market. We think that there is a big world out there where people want to be able to run on infrastructure that is not US-owned and operated because they are not US companies. Many people are aware who are users of Anthropic: Fable was shut off, or Mythos, depending on whatever name you want to give it, was shut off. That risk of you being a European company and all of a sudden having a US company pull the plug on you is changing how people talk about sovereignty. It's now no longer about financial sovereignty. And we have been very focused on the international markets from early on. Our Exion investment is very much about technical sovereignty, private cloud technology, and very secure cloud technology especially at the last mile for companies that want to run their own models and are not running frontier models. I think we are going to see a very nice convergence between the Exion model and the hyperscaler model over time. That will be a real home run for the investors.

Great. Well, thank you very much. Always great to be on. And let's all stay very focused on getting Bitcoin where we want it to get.