MARA CEO Deep Interview: Power Has Become a Core Asset in the AI Era, and Bitcoin Miners’ “Energy Transition” Is Urgent

Podcast: Natalie Brunell

Compiled by: Wu Says Blockchain

In an interview with Natalie Brunell on July 23, 2026, MARA CEO Fred Thiel said that as AI compute demand continues to grow, electricity has become the core resource being fought over by Bitcoin mining companies, cloud service providers, chip companies, and large model enterprises. He said MARA is gradually shifting from being a single Bitcoin mining company to becoming an energy platform that controls land, electricity, and data center infrastructure. He believes Bitcoin mining will not disappear, but in the future it will make more use of excess electricity generated by data center debugging, microgrids, and renewable energy. Bitcoin itself is still an important store-of-value asset, but because it lacks yield and has significant price volatility, its development window as an exchange medium may already be over. High-frequency trading and machine payments are more likely to be handled by stablecoins.

How to Become the CEO of MARA, the World’s Largest Public Bitcoin Miner

Fred Thiel: I’ve worked in the tech industry for 40 years. I started out as a software programmer, but I wasn’t very good at it, so the company moved me into product and sales roles, where I did much better. After that, I’ve worked in almost every sub-segment of the tech industry. Around age 40, I was lucky enough to lead an IoT company to go public, and then I spent some time working in private equity and venture capital.

Later, I was basically retired. I mainly served as a director for a number of companies, and provided consulting for tech company acquisitions by large private equity firms. Around that time, someone started talking to me about something called Bitcoin. I remember thinking, “Bitcoin?” I recall that earlier in my career I worked in fintech, but back then fintech was mostly bank software.

I thought Bitcoin really could solve a lot of problems, but at that time it wasn’t regulated and it didn’t have a license, so later I temporarily forgot about it. From 2014 to 2015, Bitcoin came back into my view again. I had once worked with a CTO who put everything he owned into Bitcoin—what I mean is truly everything; every last dollar went in.

He was very anti-government and completely opposed to any form of centralization. He told me, “I’m putting all my funds into Bitcoin—that’s what I’m going to do next.” At the time, he was also developing a lot of software that made heavy use of the Bitcoin blockchain. I really respected that person, so I began to think that maybe this really could be interesting.

After that, a series of coincidences happened one after another. Suddenly, someone told me, “We’ve run into a lot of problems when trading Bitcoin. The real difficulty is how to transfer funds into the exchange.” I naturally enjoy solving problems, so I started thinking about it seriously. In 2015, I partnered with a group of people to build an exchange. You can think of it as a wholesale exchange: users could first deposit funds into an account, and then we would store those funds separately across multiple different exchanges.

In that way, because we didn’t need to move funds frequently, we could trade directly and take advantage of arbitrage opportunities between different exchanges. The problem was how to obtain licenses. We put a lot of time into it—almost a year in total—and in the end we obtained a license in Liechtenstein.

Why Liechtenstein? Because Liechtenstein doesn’t have a stock exchange. The local regulators treated Bitcoin and ETH—back then, in practice, there were actually only these two types of coins worth trading—as foreign exchange currencies. That’s why we were able to obtain a foreign exchange trading license there and conduct trading.

But the problem was that since we held funds on behalf of clients, we then also needed to obtain a banking license. So this entire business model ultimately couldn’t keep running. At the same time, a friend of mine reached out and said, “Listen, someone invited me to take over a public company. We’re planning to transform it into a Bitcoin mining company. I know you understand crypto—would you be willing to join the board and help us do this?”

So I joined. That was in late 2017 to early 2018. This company was basically what later became MARA; at the time, it was called Marathon Patent Group, a so-called “patent troll” company. MARA, or Marathon back then, has a very interesting history. When it was first established, it was a vanadium mining company—an actual metallurgy and mining business.

Later, the company also did oil and gas for a period of time, tried a few other directions, and ultimately started acquiring patents. One of the patents they acquired was a foundational patent that used a limited voice command set to control mobile devices. You can think of it as similar to Siri or Alexa.

The company later sued Amazon and Apple for patent infringement. Apple ultimately paid a patent licensing fee, but the company then put all that money back into the lawsuit against Amazon. Even to this day, that lawsuit hasn’t been resolved. In short, MARA at that time was essentially a patent troll company. Like many public shell companies from that era, it was easy for a group of people to gain control of it—and later, indeed, a group of people did.

That group was the same team that originally pushed Riot to go public. They saw what they had accomplished with Riot, so they decided to replicate the same model with MARA. They invited my friend to run the company, and he then invited me to join the board.

After that, we had to first resolve various issues related to company ownership, including some toxic convertible notes. The company’s balance sheet also had problems, and dealing with all of that took some time. Then we began Bitcoin mining. The initial mining site was in Canada, and it was a hosting facility.

There’s a really great video on our website called “The Road to 23 EH/s.” You can see the mining facility in it—back then, the paint on the walls had already started peeling. Once the mining rigs started up, all the fans generated a huge negative pressure in the room, directly sucking the paint off the walls and into the mining machines. In the end, that batch of mining rigs could only all be scrapped.

These were the impressive lessons from the early days. After that, the business developed in fits and starts, and eventually we got the mining operations running. In April 2021, I left the board and formally became CEO. At that time, I was the company’s fifth employee.

From April 2021 to the end of 2023, we spent a lot of time raising capital. That was a relatively easier period for financing Bitcoin mining companies, especially for already public companies. We raised tens of billions of dollars, and when mining rig supply was tight, we placed very large orders with Bitmain, allowing the company to expand rapidly.

At that time, we didn’t invest in building or owning any mining sites—we fully adopted a hosting model, taking the light-asset approach. This made our expansion speed faster than any other company. Because in the traditional model, you need to put 20% to 30% of your capital into the mining site itself, and only the remaining funds are used to buy mining rigs. Also, building a mining site typically takes 12 to 18 months.

Our approach, instead, was to directly find partners that had available capacity, then tell them, “We have capital and compute equipment; you have electricity and site capacity—so let’s expand together.” Therefore, by the end of 2023, we had become one of the largest Bitcoin mining companies in the world.

Why “miners must control electricity”?

From the end of 2023 to early 2024, we began acquiring the mining sites that previously hosted our mining rigs, at acquisition prices lower than the reset costs of those facilities. By the end of 2024, we owned about 70% of the infrastructure used for in-house operations. After that, we started thinking about how to directly control electricity, because electricity would become a scarce resource.

I remember that in July 2021, I gave a talk at the Mining Disrupt event. At the time, I put forward a viewpoint that shocked a lot of people: you either have to become an electricity company, or you have to partner with an electricity company. Because by 2028 to 2032, with multiple rounds of Bitcoin halvings, you must control electricity resources. Electricity is the main input cost in the mining business.

Back then, almost everyone mocked this view. But today, it’s clearly electricity that has become the most critical resource. That’s also why we and many peers started transitioning: if electricity is used for AI, the revenue that each electron can generate is far higher than using it for Bitcoin mining.

That said, we are still one of the largest Bitcoin mining companies in the world, and we will continue to mine Bitcoin. Even in a world centered on data centers, Bitcoin mining is still an excellent way to optimize electricity resources. In some regions with free energy or low-cost energy, using Bitcoin mining to absorb electricity still has very good application scenarios. There are still many such localized opportunities globally. You just need to be willing to look for them, and be able to partner with people who control those electricity resources.

In English there’s a saying called NIMBY, which means “not in my backyard.” When we took over a mining facility in Granbury, Texas, on the first day we owned that facility, people showed up with signs outside protesting, saying, “Don’t do Bitcoin mining anymore.” So this isn’t something new.

The new change is that people are starting to believe that AI and data centers will drive up electricity prices and consume scarce resources such as water, while also affecting the landscape and increasing manufacturing noise. People are concerned about these things. In a sense, fear comes from a lack of information. If you already know what the end result will be, you can take appropriate actions. But if you don’t even understand what a data center is, and someone tells you, “It’s going to be loud and cause all sorts of problems,” then of course you’ll be afraid.

I think the whole industry is currently dealing with these issues. But at the root, it’s still an energy and electricity problem. Building a new power plant usually takes six to seven years. If it’s a traditional thermal power, natural gas, or nuclear power project, then it’s even longer—possibly 20 years or 30 years. With small modular reactors, or SMRs, we may be able to shorten the construction cycle to five years in the future.

But if people already can’t accept building a data center near their homes, then imagine how much they would oppose building a nuclear reactor near their homes. So I believe SMRs are more likely to be built on land owned by the government. They may also be operated by indigenous communities or tribes running data centers, and they might be more willing to deploy SMRs on their own land.

Natalie Brunell: Can you explain to an audience that’s not familiar with SMRs what they actually are?

Fred Thiel: Sorry, SMR stands for small modular reactors. For many years, the U.S. Navy has been using nuclear power to power its fleet. Traditional nuclear power plants in the conventional utility sector are usually separately designed projects. Each nuclear power plant is custom-designed, and all components have to be manufactured specifically.

But in the nuclear-powered naval system, standardized components are used. Now, some commercial companies have been founded by different teams and supported by investors including Bill Gates. Their business model is essentially to copy what Ford did for the automotive industry into the nuclear reactor space.

That means using assembly-line production to reduce cost, standardize design, and apply this model to nuclear reactors. In this way, you can build small nuclear reactors that don’t require water cooling. These reactors can also consume the old nuclear fuel that traditional nuclear reactors have already used, so there’s no need to keep buying more uranium, and it also helps address the problem of how to handle spent fuel rods.

Because they don’t use water cooling and they have a fail-safety protection mechanism—if something abnormal happens, the reactor automatically shuts down—the safety is very high. The U.S. Navy has used this kind of reactor for decades, with very few accidents, because they designed and optimized the technology to a very mature and reliable level. I believe SMRs will ultimately become an excellent solution to the energy problem. People just need to be willing to accept nuclear power first.

After the Fukushima nuclear accident in Japan and the Three Mile Island accident in the U.S., public acceptance of nuclear power has been an ongoing issue. In many cases, public perception itself decides everything. But if you look at the real situation in the U.S., you’ll find that the U.S. has enough electricity and can meet AI, residential, and industrial electricity demand.

The real problem isn’t that the total amount of electricity is insufficient; it’s what time of day that electricity can be used. A study Duke University released last year found that the total electricity demand in the AI industry is about 40 GW, while the U.S. actually has about 72 GW of electricity available, provided that data centers can reduce load for less than 2% of the time across the year.

Just think about it: 2% over 24 hours is a very short window. Data centers can use backup generators during those periods to stay operational. Historically, data center operators have wanted to locate facilities within 100 miles of so-called “NFL cities,” meaning close to major metro areas.

They also want “five nines” availability, meaning 99.999% uptime. That implies data centers must have redundant infrastructure such as electricity. But because available electricity is gradually running out, people are truly exploring and innovating. They’ll ask, “How can we become more like Bitcoin miners? How can we use energy in a way that proactively reduces load when needed, turning data centers into flexible loads? How can we make use of excess electricity already existing in the power grid?”

Remember: building a new natural gas power plant takes six to seven years. Even if Google says it will build a data center and deploy generation facilities behind the meter itself, it still needs to purchase gas turbines, complete construction, and obtain permits. All of those things take time.

Therefore, electricity that’s available immediately on the market is very limited, but the demand to chase that electricity far exceeds supply. Many people don’t realize that this electricity constraint affects not only data centers, but also all the equipment inside them, and the entire supply chain behind them.

Why did the AI compute race ultimately turn into a battle for electricity?

Fred Thiel: Suppose existing electricity can only support building 10 GW of data centers. Then the market can only absorb 10 GW worth of compute equipment, because any equipment beyond that can’t be powered on at all—can’t be connected. This would limit the potential sales volume of NVIDIA chips. At the same time, NVIDIA is also competing in chips and compute with companies like Google and AMD.

So chip manufacturers are also starting to say, “We have to lock in electricity in advance, or we’ll lose market share.” What’s happening right now is a very interesting competition. It’s not only ultra-large-scale cloud service providers that need compute and data center capacity—everyone is competing too—but also chip suppliers and frontier model companies.

For example, OpenAI and Anthropic are competing with each other. If Anthropic can’t get enough compute, then the token usage price would have to rise substantially. That would suddenly make open-source models much more attractive. So now there are multiple wars happening at the same time: at the token cost layer, it’s competition between frontier closed models and open-source/open-weight models; and there are also chip wars and wars between ultra-large-scale cloud service providers.

Meanwhile, those who control land and electricity can just sit back and say, “I’m in a very favorable position right now.” Over the past few years, MARA has accumulated a lot of relevant resources. Currently, the electricity scale we operate is about 1.1 GW. If we’re willing to expand existing sites, capacity can grow to over 2 GW.

Add to that the Long Ridge transaction we announced earlier this year, which includes a 505 MW gas power plant. Long Ridge also has more than 1,600 acres of land. We can build a large-scale data center there without needing to stop supplying power to the grid, because we can continue increasing the power plant’s capacity.

We also control a substation, so we can connect to even more electricity from the grid. In addition, we just announced a project we acquired from HIF. That project has up to about 2 GW of electricity capacity and a large campus adjacent to a nuclear power plant in Texas. The campus connects to approximately 13 to 15 high-voltage transmission lines, and those lines can withstand harsh weather and hurricanes.

The facility was originally designed for industrial applications such as power-to-fuel, but it’s actually a very ideal location for a data center campus. We’re very excited about it. So our currently controlled electricity resources are already more than 4 GW.

Does shifting to AI mean MARA will shut down Bitcoin mining rigs?

Natalie Brunell: We’ve recently heard a lot of news about Bitcoin mining companies turning to AI. Does that mean that, because Bitcoin mining profits are lower than providing electricity to AI companies, you would truly shut down your Bitcoin mining rigs? And what impact would that have on the Bitcoin network?

Fred Thiel: First, this isn’t a transformation that can happen overnight. Building a data center takes 18 to 24 months. At the same time, when we design the relevant deals, we’ll make sure that we can keep mining Bitcoin until the electricity truly needs to shift over to data centers. We’ll partition the sites in a reasonable way so that Bitcoin mining can continue operating.

In some cases, we’ll also keep a certain scale of Bitcoin mining capacity at a site, because data center electricity demand fluctuates depending on the specific use. Is it for model training or for inference? Is it supporting critical IT loads? Different scenarios have different load curves.

One advantage of Bitcoin mining is that we use a container-style deployment model, so we can truly move equipment from one site to another. For example, the HIF project we just acquired in Texas has 2 GW of electricity capacity, and local energy prices are still enough to support profitable Bitcoin mining.

In theory, we could convert all our current mining sites into AI data centers, then move all the Bitcoin mining rigs to that site, and the mining business would still remain profitable. This doesn’t even factor in any future AI business that might be deployed at that site.

Right now, we operate about 1.1 GW worth of Bitcoin mining load. Even if we migrate all of that to the HIF project, there would still be about 900 MW of capacity remaining there, and all of our other existing assets could be converted into AI infrastructure.

But as I said, this process takes years. Even if we’ve already signed leases with tenants, completing the data center construction still takes 18 to 24 months.

Is AI pulling money and attention out of the Bitcoin market?

Fred Thiel: My hair is already gray enough, and the early days of the internet were a time when I was extremely active in my career. At the time, I worked in data communications—think Ethernet-related fields. When the internet infrastructure began being built, the market looked like it had nearly infinite demand for products such as routers, Ethernet switches, networking equipment, fiber optic cables, and data centers. But later, some of that demand never really materialized.

Today, the situation is different because we’ve seen the market’s actual demand for AI compute is growing. I think the real turning point that made people realize this was when Anthropic launched Claude Cowork, and products like OpenClaw demonstrated real agent capabilities.

Around February this year, average consumers suddenly started realizing what AI actually is. The investment community also started paying attention and realized: if demand grows to that level while supply is still at such a low level, someone will definitely make a lot of money from it.

So investors started analyzing the entire value chain built around AI. NVIDIA CEO Jensen Huang proposed a very interesting pyramid model. At the bottom of the pyramid is electricity; above that are chips and compute, which is NVIDIA’s business; then comes infrastructure, including buildings, cooling systems, and so on; and at the top are the various software layers.

There’s only one factor limiting the development of all the layers above it: electricity. That’s why Bitcoin mining companies suddenly became very attractive, because they could potentially become providers of electricity resources and data center site capacity. From an economic cost perspective, building a Bitcoin mining site—including infrastructure and compute equipment—costs about $1 million per megawatt in total.

But building an AI data center requires, for just infrastructure costs and excluding compute equipment, $10 million to $15 million per megawatt. This depends on the data center’s scale, and then you still have to continue deploying compute equipment on top of that infrastructure. So it’s a highly capital-intensive business.

In the past, Bitcoin mining companies had difficulty getting credit from banks. Because banks viewed Bitcoin as insufficiently regulated, Bitcoin itself had no clear intrinsic value, and its price was highly volatile. Banks also didn’t want to enter the so-called “loan-to-own” model, because they didn’t want to end up holding mining rigs in the event of a borrower default.

That kind of thing happened during the last market trough in 2022. Many lenders ultimately had to take over Bitcoin mining equipment. But in the AI space, customers typically have investment-grade credit. If Google signs a contract with you, money flows toward you. Usually it’s a lease with a term of up to 15 years, generating recurring revenue—so to financial institutions, the risk is much lower.

Private credit providers would also be interested. So even though you need a lot of capital, you can still obtain that capital. That brings the whole capital market into the picture. And then look at individual investors. In the past, the main force driving up valuations of many Bitcoin-related public companies was retail investors.

But now retail investors see: “Bitcoin price is falling, while AI-related assets are rising—so I’ll move money over.” That’s why sell pressure started to appear on Bitcoin and Bitcoin concept stocks. Investors first moved money to gold, then to AI. Now they’ve started withdrawing from gold again and increasing their positions in AI.

I think SpaceX’s IPO sent a good warning to the whole world: not all assets keep going up forever. Companies that IPO only after becoming very mature stages often have a large number of early investors who are hoping to exit. Those investors have held shares for eight to ten years, so they naturally want to realize profits and move to the next opportunity rather than continuing to hold long term.

So I think the IPO market will change significantly in the future. Public companies will be more mature businesses, and the number of young companies will be clearly lower. At the same time, competition in the AI industry will become increasingly intense, both at the model layer and at the compute layer.

But there’s one market that will remain strongly protected for a long time: electricity, land, and infrastructure capacity. If you summarize the key constraints as three things—capital, capacity, and compute—then capacity is represented by data centers.

If you have electricity that’s available immediately, can build a data center on your own site, and can get it operational by the end of 2027 or by the end of 2028, then you’ll be in a very favorable position.

Is AI a bubble? How long can the electricity bottleneck support the industry?

Fred Thiel: I think some segments of the market really might have bubbles—for example, memory chips. Most of the show’s audience is probably too young and hasn’t experienced the personal computer wave of the late 1980s to early 1990s. Back then, as PCs truly began to become widespread, memory became a very important issue.

Suddenly, all personal computers needed memory. Every time Intel’s processors went from 286 to 386, 486, and then Pentium, the memory capacity those computers required increased even further. So the memory industry has always had very clear boom-and-bust cycles. You can look back at Micron’s stock price performance since the 1980s and you’ll see it kept rising, then falling, then rising again, and then falling again.

The reason is that when market demand exceeds capacity, memory manufacturers can enjoy high prices and high profits for as long as possible. Eventually, a company builds a new factory, capacity begins to increase, and supply gradually catches up to demand and then surpasses it. Then the market bottoms out and crashes.

After that, a new wave of technology appears, and demand starts growing again. But manufacturers stay very restrained; they don’t want to expand capacity quickly, because they don’t want to create another supply glut and industry crash. However, as demand continues to grow, they eventually still have to build more capacity.

What we’re seeing today is that South Korean memory chip manufacturers are earning huge profits, while Chinese companies are building memory chip factories at the fastest pace. This basically means that within the next 18 to 24 months, the memory chip market is very likely to enter a downward cycle again.

The storage industry, including hard drives and flash memory, experienced similar situations in the past. These cycles keep repeating. But currently, there’s another limiting factor: electricity. The electricity constraint limits how large the entire market can expand. If infrastructure capacity could be supplied infinitely, the AI market could grow faster and price wars would happen earlier and be more intense.

Because capacity is limited, the market is kept under a kind of artificial constraint, allowing costs and profits to remain high for a period of time. I believe the electricity bottleneck will only truly start to ease after at least four to five years.

What is truly difficult about building an AI data center?

Fred Thiel: It depends on where you’re starting from. If you were originally a Bitcoin mining company, then you already have land and electricity, and the electricity is already connected. That basically saves you two to three years. Next comes licensing and approvals. But first, you need to find tenants. Because if you don’t know how the project will be built, you can’t apply for permits, and the detailed construction plan depends heavily on the tenant’s needs.

Unless you’re planning to adopt a bare-metal cloud model like IREN—that is, build the data center first, deploy GPUs, and then rent the GPU compute capacity to other customers—in which case you may not need to lock in tenants upfront. But if you use a more traditional ultra-large-scale cloud service leasing model, you must identify tenants first.

The entire leasing process could take six months. During that time, you need to complete things like site inspections and detailed design work. After those are done, only then can you submit licensing applications. Depending on the region where the project is located, approval could take 90 days or possibly as long as two years. Then comes the construction phase, which averages about 18 months. The timeline could be longer or shorter depending on the project scale and location.

Labor shortages are also a challenge. At the same time, you also have to order equipment on time to ensure that facilities such as transformers, distribution units, racks, and other components arrive as scheduled. This also explains why we decided to enter the AI market through a partner like Starwood rather than pushing forward on our own. If you rely entirely on yourself, you would need to hire people who know how to persuade ultra-large-scale cloud service providers to sign with you. You’d also need to hire people who understand how to design and build data centers. And you’d need a team that can manage contractor processes, EPC engineering, and other engineering work.

Once the data center is built and in operation, you also need to know how to complete facility stabilization and commissioning, ensuring that the ultra-large-scale cloud service provider is willing to accept the project and confirm that it has been delivered. And all these tasks must be completed under very tight deadlines, because if you can’t deliver on time, you could face extremely strict breach penalties.

We intentionally delayed entering this market because we believe we don’t have clear advantages in all of the areas above. We want to find a partner who can leverage their relationships with data center owners, operators, and ultra-large-scale cloud service providers. At the same time, by leveraging their own EPC, construction, and design capabilities, they can build these projects quickly and bring in strong credit support for the projects. Starwood fully meets our requirements for a partner like this.

Natalie Brunell: I was about to ask—do you regret not pivoting earlier? After all, we’ve seen some public mining companies potentially start shifting toward AI as early as a few years ago. For you, not pivoting earlier isn’t something you regret?

Fred Thiel: Hindsight always makes things easy. Of course I could say, “If we did it a year earlier, what would it have done to the stock price?” So there’s no such thing as a perfect timing. But I’m very confident that luck can be created by yourself, because what people call luck is essentially when opportunity meets preparation. And back then, we weren’t prepared.

I’ve also said publicly many times before: if I had gone directly to Google back then and said, “I want you to be the tenant of the data centers we’re planning to build,” they would probably have thrown me out of the office. You’re just a Bitcoin mining company—you don’t know how to build a Tier IV data center, right?

So we must work with partners who truly know how to get this done. Starwood is excellent. They’ve built about 7 GW of data centers for companies like Google, Amazon, and Microsoft. They know what they’re doing, and the cooperation is currently going very smoothly.

If you also look at the financial arrangements between both sides, you’ll see that this model is very favorable for us, because it allows us to participate in projects in a very light-asset way. That means we don’t need to issue a huge amount of stock to fund these projects, so we don’t have to overly dilute existing shareholders’ equity.

At the same time, a lot of debt financing is available in the market. Even if you calculate it using an 80% loan-to-value ratio, this model is still very light on capital for us. So from a capital efficiency perspective, this approach is very efficient, and it also allows us to develop larger-scale capacity. That’s why we’ll keep focusing on what we’re best at: acquiring land and electricity assets, and then turning those assets into the highest-value use cases.

Why did MARA sell about 20k BTC?

Natalie Brunell: What prompted you to make that decision in the first place? We often see that Bitcoin miners sell immediately after mining Bitcoin. The operating environment is also difficult now—because of Bitcoin price pressure, many public and private miners are actually mining at a loss.

Fred Thiel: We’ve always been very clear that we’re not a digital asset treasury company. For us, holding Bitcoin has always been just a treasury management strategy—basically a way to hold cash. In fact, our first Bitcoin purchase was in January 2021. At the time, we went directly to the market to buy Bitcoin. We had $150 million in cash on the company balance sheet, but since there was no more mining rig capacity available in the market to purchase, we couldn’t use that money to buy mining rigs.

So we decided to allocate funds into Bitcoin as a way to store cash, because we expected the Bitcoin price to go up. Looking back, in January 2021, the Bitcoin price was around $15k, so that was obviously a very correct decision. Now look at last year’s third quarter. I remember saying in the company’s public earnings call that the market price was already getting somewhat overheated—and that was indeed the case.

At the time, we held about 55k BTC. At a price of over $100k per Bitcoin, the value of Bitcoin on the company’s balance sheet was roughly $5 billion to $6 billion. So yes, holding Bitcoin was indeed a very successful strategy.

But as the price of Bitcoin fell, we were also carrying about $3 billion of convertible debt. About $1 billion of it might mature in 2027 because investors would exercise their put/call rights—these are structured features. Our general approach has been that when the trading price of convertible bonds falls below par, it’s usually reasonable to repurchase those bonds.

So we sold about 20k BTC to repay about $1 billion of convertible debt. Given that those convertible bonds were trading at a discount at the time, economically this transaction was us selling Bitcoin at about $80k per BTC—which was very reasonable. Of course, it also sent a signal to the market. But not long after that, Michael Saylor did the exact same thing.

Does selling Bitcoin mean MARA no longer has faith in Bitcoin?

Fred Thiel: Because the market has indeed changed, and the pace of change has been extremely fast. Honestly, I remember that during interviews around the election, Bitcoin was constantly hitting new all-time highs, and everyone was preparing for Bitcoin to go to $200k or even $300k. Then suddenly things started to change. A lot of Bitcoin treasury companies appeared, and now—about one and a half years has passed, nearly two years.

Some companies started selling Bitcoin, their original strategy changed, and mining companies shifted toward AI. Even people who strongly support Bitcoin and stay at home can sometimes feel confused when faced with this information.

Fred Thiel: Yes. Gold as an asset class has been around for a long time. There are some periods when gold is extremely popular, and there are also periods when the market doesn’t really favor gold at all.

Recently, we’ve also seen gold have its own spotlight moment.

Natalie Brunell: The two asset classes that performed worst this year are gold and Bitcoin.

Why hasn’t Bitcoin become a widely used payment currency?

Fred Thiel: So I think this needs to be looked at over a longer time horizon. But as an asset, Bitcoin has a fundamental challenge: it doesn’t generate yield. Therefore, as a store-of-value instrument, Bitcoin’s price depends entirely on whether more people want to hold it than want to sell it. It’s essentially an asset determined by supply and demand.

When Satoshi Nakamoto originally wrote the white paper, Bitcoin was designed as a currency. But what’s worth discussing now is the overall security budget of the Bitcoin network—especially given that so many mining companies are starting to shift to other businesses. Under the original design, as block rewards keep halving, transaction fees should gradually become an important source for compensating miners. But in reality, that hasn’t happened.

Natalie Brunell: Right. Transaction fees are now close to historical lows, aren’t they?

Fred Thiel: Yes. Transaction fees for block space are very low now. I remember it might be around 5 basis points. Ultimately, the mining rewards themselves are also already very low. The block subsidy compared with the past has been significantly reduced. The good side is that the number of new Bitcoins produced by miners and entering the market has decreased. But at the same time, market demand for Bitcoin is also falling.

Looking back at the period from around the U.S. election through 2025, the market first saw Bitcoin ETFs, and then it saw digital asset treasury companies. These factors all created new demand, while Bitcoin’s supply didn’t change.

So Bitcoin prices kept rising—surpassing the level that natural demand growth could support. That’s why we said in last year’s third quarter that the Bitcoin price was somewhat overheated and needed to cool down closer to the statistical mean. Based on that view, Bitcoin’s more reasonable price right now might be around $90k.

After a price surge this large, it typically needs to go through a full drawdown for it to return closer to the long-term mean before it can start rising again. So I think this process still needs some time. However, during conflicts or turbulence, Bitcoin remains attractive. For people who want to move assets quickly, it has very high value.

As for future business uses of cryptocurrencies such as AI, I believe most transactions will ultimately be executed via stablecoins. When two counterparties trade, both sides want to clearly understand the value of the exchange medium.

If you exchange eggs for milk, both sides may assign different valuations to the two commodities. But if you exchange dollars for a dollar-denominated product, both sides know exactly what it’s worth. Dollars, of course, are also affected by factors like inflation. But if you need to complete thousands of transactions per second, even small fluctuations in Bitcoin’s price could immediately change the actual value you receive. Especially when profit per trade keeps getting lower while transaction frequency keeps getting higher, this kind of volatility becomes a very serious problem. So I think Bitcoin’s path as an exchange medium is unfortunate—it may have already missed its best development window.

But it’s still an excellent store of value. If you want to keep your assets outside centralized control, or want to hold assets in a form that can be convenient to transfer in the end, then Bitcoin still has value. As long as centralized institutions haven’t completely cut off the channel to convert Bitcoin into cash, Bitcoin will still be an asset held by specific groups of people, giving their wallets and wealth maximum flexibility.

MARA currently doesn’t plan to keep accumulating more Bitcoin.

Fred Thiel: Yes, MARA currently has no such plan. As you just said, most mining companies are now mining at a loss, so continuing to accumulate Bitcoin doesn’t make sense. Because then we would have to raise cash somewhere else to sustain operations, and we don’t want to do that by issuing stock and diluting shareholders’ equity.

Natalie Brunell: Compared with five years ago, are you less bullish on Bitcoin now?

Fred Thiel: No, I haven’t reduced my confidence in Bitcoin. I’ve just placed it in a different position. I owe fiduciary responsibility to shareholders, and I must generate the best possible returns on the capital they put in. We have an asset—we have land and electricity resources. I can convert those resources into AI infrastructure, and thereby earn returns far higher than Bitcoin mining. So I must ensure the company focuses on realizing this opportunity.

We have many stakeholders, including employees, shareholders, and industry partners, and we need to do our best to create value for them. I think in the long run—if I remember correctly, I also talked about this at another Mining Disrupt event—Bitcoin mining will ultimately make more use of excess electricity that would otherwise be wasted in day-to-day economic activities.

At that time, mining won’t be an extra cost; it will happen naturally. There will be many micro-mining activities around the world. Large industrial mining companies will be viewed more as a tool for balancing electricity loads, or as a controllable electricity sink. In addition to obtaining Bitcoin rewards, Bitcoin mining itself also has practical value at the electricity grid system layer.

For example, when you stabilize and commission a data center and gradually bring electricity online, you typically use load banks to consume that electricity. But you could just as easily use Bitcoin mining rigs instead of load banks.

The cost of the two approaches is the same. But from the perspective of Bitcoin mining, it’s like mining at zero marginal cost. So eventually, Bitcoin mining will naturally migrate to the scenarios where it fits best. These applications will ensure that the network always retains enough compute so that Bitcoin can continue to run securely and circulate. After that, the only question left is whether the Bitcoin community will continue to maintain the code and avoid the entire technical system becoming overly rigid.

Can Bitcoin still grow into an asset worth tens of trillions of dollars?

Natalie Brunell: It sounds like you’re describing a future where Bitcoin remains a relatively smaller asset class rather than reaching the scale some people in the community predict—$200T, $50T, or even $100T.

Fred Thiel: I think Bitcoin can grow into an asset class worth $2T, and achieving that by itself is already a very significant accomplishment. It’s already in the global top 10 asset categories.

Things like this take time. The Bitcoin ETF market achieved growth in just one year, which is what took the gold ETF market two to three decades to achieve. But to maintain stable long-term growth, it must be built on a solid foundation. I think Bitcoin is gradually forming that foundation now.

But my generation is still largely constrained by the traditional financial system. Look at my kids: they were exposed to and invested in Bitcoin very early. But as they grew older and started families, they started considering index funds and traditional investment tools, and gradually allocating across multiple asset classes.

By the next generation—my grandchildren after they grow up—Bitcoin may just be one of many assets they invest in. They won’t treat it with the same intense enthusiasm as the generation that originally built Bitcoin.

Natalie Brunell: Why did you change your positioning on Bitcoin? Over the past roughly five years, what happened that made you feel things diverged from the original direction? Why hasn’t Bitcoin become a broader store-of-value tool, and why hasn’t it replaced some asset classes and platforms in the traditional financial system?

Fred Thiel: The monetary system is controlled by large banks and governments. If governments in each country treated Bitcoin as a national reserve asset—if banks and regulators allowed banks to hold Bitcoin easily on their balance sheets, and allowed banks to use Bitcoin for interbank settlement—then things could be completely different.

Bitcoin is actually very suitable as a settlement currency, because interbank settlement doesn’t require processing millions of transactions every minute; usually it only processes a small number of large-value transactions. For scenarios like balance-of-payments settlement, Bitcoin is also a very good solution. But those traditional institutions that control money, the financial system, and the rules simply haven’t allowed Bitcoin to truly enter that system.

Natalie Brunell: Do you think that won’t happen in the future either? Doesn’t it have to be a gradual evolution?

For example, the U.S. might establish a Bitcoin reserve in the future. There are even views that the U.S. government has already started mining Bitcoin—we can also see some on-chain hints that might relate to that.

Fred Thiel: I think some government departments are indeed trying. If they have excess electricity, they’ll look into what they can do with it. But the question is how to turn experiments and innovation into large-scale, real-world applications.

At the end of the day, Bitcoin has to have sustained demand. Only when more people want to buy Bitcoin than want to sell it will the price rise. This creates a self-reinforcing virtuous cycle: when Bitcoin’s price rises, more people want to hold it; as the price keeps rising, people start viewing it as an even better store of value.

Especially when fiat currencies keep losing value due to inflation, Bitcoin may still preserve its own value. So I think Bitcoin is similar to gold. Gold has theoretically always been viewed as a perfect inflation hedge, but it also has its own cycles and seasons.

Right now, Bitcoin may be in its fall-winter phase. But spring will eventually come again, just like it has happened before.

What is MARA’s development roadmap looking ahead five years?

Fred Thiel: I think we’ll always keep mining Bitcoin to some extent. In two more years, Bitcoin will have its next halving; in four more years, another halving. When the subsidy per block ultimately falls to less than 1 Bitcoin, and if electricity can be used for AI, then the cost of mining Bitcoin will become quite high.

I still think Bitcoin mining will gradually become part of certain computing devices, and may be integrated into devices such as air conditioners, or built into many products that can use idle electricity. For example, solar panels. I expect that over the next few years there will be more and more microgrids, because consumers want to control their own energy—or at least directly use the electricity they generate.

People like us who used to live on the U.S. West Coast have become quite familiar with solar energy. While utility companies have messed things up when designing solar pricing and billing mechanisms, in theory the way to solve grid balancing is to install solar panels on a large number of buildings and deploy batteries inside those buildings.

When the grid needs electricity, it can draw on the energy stored in those batteries, creating a more balanced grid. But when the batteries are already full and the solar panels keep generating power, what do you do with the excess electricity? You can use it to mine Bitcoin.

Two years ago, we made a small investment in a company. Its business model was to install solar panels on residential rooftops, paired with Bitcoin mining rigs. When the batteries don’t need charging and the house isn’t consuming that electricity, the system automatically uses excess power to mine Bitcoin. It’s a highly capital-intensive business, because the company has to bear all the upfront costs such as the solar panels. In the end, the company went under, but the idea was very good.

I think in the future, someone will eventually integrate very simple circuits from Bitcoin mining ASICs into various devices. ASICs themselves aren’t complicated; what’s truly complex is the power management part. But if you’re not doing industrial-scale mining, power management isn’t as important for you. Then there will be lots of opportunities to mine Bitcoin using excess energy. At that time, you might not even care about mining costs, because that electricity is basically free.

Natalie Brunell: That’s definitely true. It’s hard to imagine that in just a few years, the market narrative has shifted from “Bitcoin consumes too much energy and will eventually boil the ocean” to something that has almost completely disappeared. Now similar narratives seem to be shifting toward AI. But at least in my view, the public hasn’t been worried about AI’s energy consumption heating the whole planet the way they worried about Bitcoin back then.

Maybe it’s just how I personally feel. In the past, people were extremely fearful about Bitcoin’s energy consumption, but with AI, people may choose to ignore it because everyone uses it every day.

Fred Thiel: Going back 25 years, when people talked about the internet, they said, “Oh my God, the internet only has crime and pornography—everything is bad. The internet won’t bring us any good.” Then later, people realized they couldn’t live without the internet.

AI will likely go through a similar process. I think Bitcoin will also go through similar changes in its own way. Eventually, it will become a normal thing: you’ll hold it as one of many investment assets, and use it to store value. If in the future the dollar really starts collapsing, I can assure you that demand for Bitcoin would surge dramatically.

Natalie Brunell: Do you think there is some kind of price ceiling for Bitcoin? Is there a price level where you’d feel, “I don’t think it can go any higher”?

Fred Thiel: You need to recognize that price depends on both sides of the value equation. Suppose the dollar is hit by severe inflation and loses 20% of its value each year. Then even if Bitcoin’s price denominated in dollars increases 20%, your purchasing power would remain unchanged.

So the real question is what happens to the value of global money and other assets. In periods of high inflation, hard assets tend to become extremely valuable—for example, art, land, and real estate. These tangible assets are often difficult to replicate or replace. We’ve seen these cycles repeatedly.

Gold and Bitcoin may also fall into that category. In high inflation periods, Bitcoin may become a destination for money seeking safety. Now look at the digital-native generation you’re in, and the generation my kids belong to. For them, physical gold might feel like, “How on earth do you even use it?”

Whether it’s users of Revolut, users on Reddit, or young people who got started with stock trading through memes and meme stocks—none of them would likely choose complicated gold certificate trading, or use financial instruments that don’t actually settle in gold. But they really like the concept of Bitcoin. For this group of people, the more volatile Bitcoin is, the more exciting it might be.

What is the real threat of quantum computing to Bitcoin?

Natalie Brunell: What do you think about the BIP-110 proposal and the governance debate that seems to be tearing the community apart? It currently looks like fewer than 1% of Bitcoin miners support the proposal. What’s your take?

Fred Thiel: The MARA Foundation has its own position on this. I think this debate isn’t particularly constructive for the industry as a whole. Meanwhile, the threat of quantum computing to Bitcoin wallets is real—in the “near term,” I mean roughly the next five to ten years.

This isn’t a problem with the Bitcoin ledger itself; it’s a problem with wallets, and it only involves part of the wallets. So I think there are better ways to handle quantum risk. One possible solution is to introduce a certain time delay for specific types of wallets when transactions occur.

But all of these measures require the entire community to suddenly act like some kind of central authority and coordinate collective action. That conflicts with Bitcoin community ideology, because what Bitcoin emphasizes is decentralization. My personal view—this isn’t the MARA view; it’s just my own view—is that if your Bitcoin is stored in an early, traditional wallet, then create a new wallet and move your Bitcoin there.

Don’t keep Bitcoin long term in a wallet or address that has repeatedly been used for trading. Use a new address each time, so your public key isn’t exposed for too long and you won’t really face this issue. If you’ve already forgotten the wallet’s private key, then you can’t move that Bitcoin anyway. Even if someone steals it later, it’s just bad luck.

But if one day the Bitcoin held by Satoshi suddenly starts moving, I can guarantee that people will say, “These wallets were compromised,” or “Satoshi is actually still alive.” In contrast, what I’m more worried about is what happens in the market if Satoshi’s Bitcoin starts moving.

Natalie Brunell: If those Bitcoins start moving, doesn’t that mean the quantum attack has happened?

Fred Thiel: Not necessarily. It could also simply mean that Satoshi decided to move his own Bitcoin.

Natalie Brunell: I don’t know. I do think someone might have Satoshi’s private key. Maybe Satoshi is still alive.

Fred Thiel: Yes. If you look back at Bitcoin’s very early development stages, a lot of things actually happened that most people don’t know about. I’ve spoken with many people who are making documentary films about it. I think in the next few years, different documentary makers will release some very interesting pieces telling the story of what happened in Bitcoin’s early days.

If you truly dig into the emails from back then and what happened behind the scenes, my personal belief is that Satoshi can no longer control his wallet. In fact, it’s possible that no one can control those wallets at all, because the private keys were split and stored, and later something went wrong.

Natalie Brunell: That’s possible. Then ultimately we need to decide what to do with those Bitcoins, right?

Fred Thiel: It’s a bit like the issue in the lawsuits related to Craig Wright—the key question is whether you can prove that you actually control a given wallet. Of course, there are also people who don’t want the outside world to know they hold a large amount of Bitcoin.

When it comes to quantum computing, I always tell people one thing. In an earlier career, I served as chair of one of Europe’s largest cryptography technology companies. That company produced hardware security modules, or HSMs. NATO, credit card companies, banks, and other institutions with extremely high security requirements all use devices like this to store cryptographic keys.

At the time, we also partnered with the U.S. National Institute of Standards and Technology, or NIST, to help evaluate post-quantum algorithm proposals in order to find replacements for RSA. You can think of it like this: almost all cryptographic systems used today for bank accounts, email, and website SSL certificates are built on asymmetric encryption.

That means one side needs to do heavy computation, while the other side’s computational burden is relatively light. Fundamentally, it’s an asymmetric architecture consisting of a public key and a private key. Right now, the value of the assets and information hidden behind RSA keys is far higher than the assets held in early Bitcoin wallets. Hackers may already have obtained some of that encrypted data, but they still can’t decrypt it yet.

For example, you can monitor huge amounts of network traffic from an internet service provider and save the logs. Those data are encrypted because they’re transmitted via SSL. But if you have a sufficiently powerful quantum computer, you could decrypt them. Then you can directly obtain usernames and passwords without needing brute-force attacks.

You can obtain the usernames and passwords for wallet accounts, bank accounts, securities brokerage accounts, and essentially almost all other accounts. In theory, you could even quietly transfer $0.1 from every one of Citibank’s accounts. It might take a long time before anyone notices.

By doing it quietly and in a distributed way, you can steal far more funds. By contrast, if someone transfers just one Bitcoin out of a famous early Bitcoin wallet, the whole world would immediately know.

And who is most likely to obtain the most advanced quantum computing capability first? The answer is nation-states. What would they use it for? Suppose North Korea has a quantum computer—it would first steal funds. Bitcoin would be the last place they attack. After that, they might convert the stolen assets into Bitcoin to move funds, but clearly there are many better targets and ways.

I remember very clearly sitting down with people from Microsoft and Google to discuss, “What public signals should we watch to determine that quantum computing is close enough that people truly start to worry?” The “people” here are mainly the chief security officers at banks.

The answers they gave were basically the same: you’ll know the quantum threat is imminent when Microsoft starts replacing security certificates, and Google also starts upgrading the security certificates used for websites and email to post-quantum standards.

Natalie Brunell: Very interesting. I’ve even heard some people say that when they die, they would destroy their Bitcoin to increase Bitcoin’s scarcity. But I don’t really understand that approach. Because if wallets later must be upgraded to post-quantum wallets, wouldn’t you just be leaving those private keys to hackers?

Fred Thiel: Ultimately, it may end up like that. But if your Bitcoin private keys are stored in some kind of hardware device, and you always use a new wallet address every time you move assets, then overall it’s still fairly secure.

Natalie Brunell: But if you’re already dead and a post-quantum wallet appears later, then you can’t complete the upgrade. You thought you destroyed those Bitcoins, but they might still be stolen by someone else, right?

Fred Thiel: Have you interacted with those Bitcoin investors whose net worth numbers have a “B” after them—that is, who are worth billions—and who weren’t that rich before investing in Bitcoin? Almost all of them are seriously doing estate planning. What’s interesting is that during last year’s Bitcoin price surge, there were some trading days when $1 billion worth of Bitcoin was liquidated.

Most people don’t realize that you can convert Bitcoin into a Bitcoin ETF like IBIT without necessarily treating it as a taxable sale. Why do that? First, it’s treated as an ETF. You can use it for collateralized borrowing, or include it in a family trust and estate planning.

So we’ve seen many extremely wealthy Bitcoin holders suddenly start doing estate planning. The issue is that if you die, what exactly does your spouse, your estate executor, or other related parties do—how do they access your wallet, manage private keys, and handle all sorts of other complicated matters?

So putting part of your assets into the traditional financial system still supports Bitcoin as an asset, because the underlying holding is still Bitcoin.

Natalie Brunell: To be honest, your overall view on Bitcoin sounds somewhat pessimistic.

Fred Thiel: No, I want to reemphasize that Bitcoin is undeniably a core component of my personal assets, and I’ve always been very confident in Bitcoin. Like all assets, it goes through highs and lows. I’m also very proud that I’ve been able to play a role as Bitcoin gradually moves toward maturity.

I think Bitcoin will continue to grow. It will shift from being an asset mainly held by people full of passion, to an asset that rational investors and committed believers can participate in together. I believe Bitcoin will remain long term, continue to maintain its value adjusted for inflation, and find its balance over the long run.

As for the so-called “Bitcoin to the moon”—unless the world experiences some major disaster that drives massive capital inflows—I think Bitcoin is more likely to continue as an asset that can hedge inflation and preserve purchasing power, and remain part of people’s investment portfolios. And frankly, that’s a good thing too.