The Pomp Podcast - #1333 Dan Roberts on The Future of Bitcoin & AI Data Centers

Episode Date: March 27, 2024

Dan Roberts is the Co-Founder and Co-CEO of IREN. In this conversation, we talk about data centers for both bitcoin mining and artificial intelligence, energy sources, decision making framework, market reactions, competitors, future of bitcoin, future of AI, and more. ======================= Introducing Espresso - the world’s most interactive portable display. They have a portable screen that is incredibly light, comes with a nice stand, and the user interface is very easy. Anyone who listens to this podcast can go to ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠us.espres.so/pomp⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠. They have a brand new offer waiting for you.  ======================= BetOnline.ag is a proud sponsor of the the Pomp Podcast. Use crypto to bet on sports, play poker and enjoy casino games at BetOnline. Visit https://promotions.betonline.ag/pomp and use promo code POMP100 to receive a 100% matching bonus on any crypto deposit. BetOnline boasts no crypto transaction fees, and processing is anonymous, instantaneous and secure. ======================= Pomp writes a daily letter to over 265,000+ investors about business, technology, and finance. He breaks down complex topics into easy-to-understand language while sharing opinions on various aspects of each industry. You can subscribe at https://pomp.substack.com/ ======================= View 10k+ open startup jobs: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠https://dreamstartupjob.com/⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ Enroll in my Crypto Academy: https://www.thecryptoacademy.io/

What's up everyone? This is Anthony Pompliano. Many of you know me as Pomp. You're listening to the Pomp Podcast, which is my effort to find the most interesting people in the world and sit with them for hours while I ask questions in an effort to learn. So it would mean the world to me if you would subscribe to the show on your favorite audio platform, watch episodes on YouTube, and tell your friends and family about the podcast. My goal is to help millions learn from the world's most interesting people. So let's get into today's episode. Today's episode is with Dan Roberts. He's the co-founder and co-CEO of Iren. In this conversation, we talk about data centers, both for Bitcoin mining and for artificial intelligence

compute. Dan and his team have built an incredibly interesting business. They built those data centers using cheap renewable energy sources that are usually outside of areas where other people want to gather that energy. And then they're serving both Bitcoin miners and AI companies. This new era of all sorts of advanced computation is only beginning, and companies like Iron seem to be best poised to capitalize on it. Dan and I get into the decision-making framework when they decide to mine Bitcoin for serve their AI customers, how the market receives the different stories, where Bitcoin's going, where AI is going, how they source GPUs, what their competitors are doing, and why artificial

intelligence has become such an attractive use case. It's all in here and Dan did a fantastic job educating me, so I appreciate his time and I hope you enjoy my conversation with Dan Roberts. Anthony Pompliano runs Pomp Investments. All views of him and the guests on his podcast are solely their opinions and do not reflect the opinions of Pomp Investments. You should not treat any opinion expressed by Pomp or his guests as a specific inducement to make a particular investment or follow a particular strategy, but only as an expression of his personal opinion. This podcast is for informational purposes only. the second I leave my desk. If I'm traveling, if I go to a coffee shop to do some work

or just want to work from the kitchen table, my laptop doesn't have enough screen space. I lose my command center and my productivity falls off a cliff. It's a major problem. But this is where Espresso comes in. They have a portable screen that is so beautiful that you think Steve Jobs came back

from the dead to create it. The thing is incredibly light. It comes with a nice stand and the user interface is so easy that I figured it out. How to do it in less than three minutes. If you listen to this podcast, you know that's not an easy feat. So the Espresso team and I, we became friends.

I got to know them because I really like the product. And those screens, they now want to offer them to any fan of the podcast. So we struck a little deal. Here's how it works. Anyone who listens to this podcast can go to us.espresso. Or, if that's too confusing, just go click the link in the description. If you go to Espresso's website, they've got a brand new offer there sitting for you. You get a little discount and you'll get a beautiful screen.

Trust me, I use mine every day. You'll love the Espresso screen and I think it'll make you more productive. Go check them out today by clicking on the link in the description. This episode is brought to you by BetOnline. Do you like making a profit from sports betting? Well, set yourself up to take home the most profit possible using crypto to fund your sports betting, casino, and poker account at betonline.ag. You can avoid costly transaction fees, get your payouts lightning fast, and do it all securely and anonymously with the highest

deposit and withdrawal limits in the industry. If you want to get in on the action, do it the smart way with crypto at betonline.ag. Head to the website, sign up with promo code POMP100 to get 100% bonus on your crypto deposit today. If you go and you deposit, they'll give you 100% bonus if you use promo code POMP100. BetOnline. The game starts here. BetOnline.ag. Go check them out today. All right, guys. Bang, bang. I've got Dan here with me. I thought a great place to start this conversation is you guys burst onto the scene. You built these data centers where you're doing Bitcoin mining. Bitcoin is surging. Everyone's really excited. But maybe you weren't actually just building a Bitcoin miner. What it seems like you guys were doing was really building

data centers where you could put Bitcoin mining or you could put artificial intelligence, computational power as well. And so maybe we just start with like, what is a data center and how did these exactly work? Yeah, look, that's right. Clearly, we didn't see the current wave of AI taking off as it did. But when we built this business five and a half years ago, it was all about the thesis that we needed power dense data centers. So what is a data center? Historically, over the last couple of decades, you've seen the emergence of these major capital city data centers, which are typically optimized for things like lifetime cloud computing. So mission critical, hospital, government, corporate cloud computing systems. And what we foreshadowed when we started Iron was that the emergence of just power dense

compute, so just workloads that needed to crunch data. Now, whether that was genomics, whether that was just gas and oil reservoir analysis, whether it was machine learning, AI, Bitcoin at the time, we were very bullish on Bitcoin. We remain bullish on Bitcoin. What a great way to bootstrap the development of these power dense data centers. And the key difference between what we've built and traditional data centers is their ability to handle these power dense workloads. And that goes to two main things. One is what they term rack density.

so the amount of power that you can physically feed into a rack which is yeah a column of computers and the other one is ventilation and cooling the ability to transfer the heat that is generated from that computing power away from the rack to keep the computers cool historically capital city data centers have grown up in this world of using around 550 kilowatts of power per rack, we've started out doing kilowatts. Now, the NVIDIA GPUs that we're using for AR require about 40 kilowatts per rack. So for us, we're arguably over-specced from a power density perspective. For traditional data centers, they're effectively having to replace two out of every three racks with empty racks in order to accomplish these NVIDIA GPUs. So it's an enormous opportunity for us.

Now, when you think of building these data centers and operating these data centers, How much of it is like the classic real estate, you know, location, location, location, even though people aren't necessarily coming by like retail foot traffic, you do have to think a lot about like energy sources. If you're not near the energy, it's probably more expensive. And so how do you think about energy sources, select those energy sources and think about locations to put these data centers? Absolutely. And look, most of our backgrounds is in infrastructure, renewable energy development, as well as the data center side. And what we saw firsthand was this Western government push to promote the build-out of renewable energy. And like most government programs, there's side effects of how that works. And one of those unintended consequences is building a lot of renewable generation in places where people cannot use it.

So you think about the Blackstones, the Black Rocks, all these people building and buying solar farms. it doesn't matter where they build them they will still get the renewable energy certificate so our Childress site which is located up in the Panhandle region of Texas there's around 32 gigawatts of wind and solar the transmission line's 12 gigawatts to actually export that load down to Dallas and Houston and the other demand centers so for us the ability to go and locate in what is a quite sizable regional town in Childress, but take up this excess renewable energy and turn it into computing power, whether that's AI, Bitcoin, or future applications, is something that's really compelling and attractive from a cost and value generation

perspective. Now, when we think about what's going inside of the data centers, some data centers cost, you know, I don't know, $8 million to build, right? It's incredibly expensive. Bitcoin miners historically have loved the fact that they can set up what people that I know in the data center business would call chicken coop. You don't need all of the redundancy. You don't need all of the super clean lifted floors and great ventilation and all these things. It was kind of like you just needed the bare minimum to get operational. And so how do you all think about maybe like the quality or the cost of the data centers that you're building? Yeah, look, I think it's fair to say we've spent a bit more than your average Bitcoin miner on these facilities, but substantially

less than traditional data centres. So we haven't done CCANs, shipping containers, old warehouses. We certainly have clean floors. We have very clean, dust-free environments. And from day one, we've built out the facilities, both from an infrastructure, networking, communications perspective to be fit for purpose for different functionality, i.e. different use cases. So we've installed ASICs historically at most of our data, all of our data centres. And recently, we've been swapping out

some of those ASICs for the NVIDIA GPUs. And I mentioned the ventilation and the power density earlier. That was fine. That was easy in the sense that we can manage the heat. But because we'd implemented all these additional redundancies around network comms, etc., we were able to plug in the GPUs and launch an AI cloud service that has been really well received by the market. I think one of the key elements of where the generative AI market is today is it's very focused on training models. So you've got a number of these startups or scale-up businesses that are raising a lot of capital, building out training models that really do look like they're going to deliver valuable use cases. And during this training phase, it's very

project-based, i.e. run your workload for three months, six months, develop the model. And in the future, you'll then have more of the inference workloads. So, the key distinction I'd like to draw is training versus inference. Training is developing the model. Inference is when you log on to chat GPT with a prompt or mid-journey for an image, type in your prompt, it generates the response lifetime. That's inference. So, for us, we're doing a lot of the training models. The next wave of AI and the next wave for us is really tapping into that inference market. Now, when I think of these two use cases, it almost seems like you guys have preserved optionality.

You could do Bitcoin mining. You could do AI compute. There are some differences, not just in the actual machines that you use. So for example, with Bitcoin miners, there's persistent demand for the mining hash rate, right? You can just hook into the network and there you go. You're off to the races. With AI, do you need a sales force?

Like, how do you go get customers? is there's so much demand that like you guys like put up a flag on the internet you're like hey we have compute and people just like storm the facility you know how how is this working right now absolutely it's a key point and in fact i think it was four years ago we signed our first mou for non-bitcoin um applications with dell and that was to bring out their customers and hardware to do high performance compute like ai in the very early phases machine learning etc and one of the reasons we aborted or postponed at that point in time was it's hard to develop all these customer base tiny contracts it just doesn't move or didn't move the needle at that

time and it was much easier to plug in these bitcoin mining machines they're generating substantial cash flow from day one with no sales force and there's always a buyer of last resort in there in the bitcoin market so today with the scale and size of where the ai markets now reached, it's much easier to win large contracts. And in fact, we're having multiple conversations with people that want substantially more capacity than what we've currently got. So that's a fantastic problem. And at the other end, which we're clearly working on. So yes, we are developing a sales force. To date, it's just been us internally

having conversations with these AI companies. And in terms of the optionality, what's fantastic is that we build these data centers. We can immediately install them with Bitcoin ASICs generating cash flow off them. And then as the AI market progresses, we've always got data center capacity available to service it. So we're not, again, like other data centers, which will only build once they've got a customer contract. You've then got to kick off a six to 18 month construction program. We've always got immediately available capacity, which is a key competitive advantage. And in terms of the optionality and value generation, we've now got this kind of call

option or exposure on two enormous exponential macro trends. And what we've been able to do is to leverage a real asset and leverage it into these exponential revenue lines because of the emergence and growth in technology. Now, what is the difference between the monetization, right? in terms of how much you get paid if you're mining Bitcoin versus AI? And maybe we can think of it on like a unit economic basis, or we could also think about it just in terms of like, maybe you get more compute into a data center or something.

Like what is the decision framework between should we mine Bitcoin or should we service, you know, AI companies? Yeah, it's a really good question, Paul. And it's something that we think a lot about. Right now, we're going very hard at both, particularly given the state of both markets. So we're scaling up to 20 exahash over the next nine months of capacity, which will take us to one of the largest listed miners. And we've been able to do that because we've got all the power, the land, we've invested ahead of the curve, and now it's just punching out the capacity.

So we're going full steam ahead of that. And then the AI side of the business is very easy to backfill into existing data center capacity as the customer demand continues to mop up the additional capacity that we bring online. I think the key reason why these strategies are not mutually exclusive is the capital intensity of GPUs. So the capital intensity of GPUs is very different to ASICs. So the decision framework is build these data centers, and then you've got the choice. Buy the Bitcoin mining machines. They're typically historically around 12 to 18-month paybacks, cash on cash. It's a bit earlier or a bit sooner in the current environment, given where Bitcoin is trading.

And then on the GPU side, they're around about two-year payback. Interesting, because in one of them, you're getting paid in dollars. Those dollars don't appreciate. So you want profit and then you got to figure out what to do with those dollars. With Bitcoin, it's appreciating now. It does draw down. And so there's like a currency risk to some degree in both directions. And so how do you think about the Bitcoin that you do mine, whether to hold it, whether to sell it for cash or sell it to reinvest in the business?

Just talk through kind of the almost like capital allocation decision when you're actually mining Bitcoin rather than receiving dollars on the AI side. Yeah. So today we have liquidated all our Bitcoin immediately. Every day we get the expected amount of Bitcoin into our regulated exchange account. We sell it for cash, withdraw that to our bank account, pay the power bill at the end of the month and pocket the rest as profit. And that's been a strategy to leverage our infrastructure, data center, real asset background and generate cash flows from this business. Now we're as bullish as anyone on Bitcoin. But what we haven't seen is worthwhile is to dilute and issue lots of shares to go and buy Bitcoin and then sit on them. We've had a pretty firm view from day one that investors

shouldn't be paying us to hold Bitcoin on their behalf. If people want to own Bitcoin, go and buy it in cold storage. If you don't want to manage cold storage, you've now got ETFs. I don't understand why people would want to pay me to hold Bitcoin on their behalf and dilute our share price to do that. Essentially, the other miners are doing. So for us, we have been reinvesting that operating cash flow to minimize dilution and hyperscale the growth in our business. Now, as we approach that 20X a hash, that's a substantial amount of capacity, a substantial amount of Bitcoin that we're generating. Do we change strategy at that point? Perhaps. Maybe we do start holding a little bit of Bitcoin. Do we start looking at dividends?

Now, there's two stories here, and you could be a Bitcoin mining company, you could be an AI company, but it sounds like almost you guys are like, no, we're both. There's like this like advanced compute story. And so talk a little bit as to conversations with investors or people in the public markets and how do they respond or receive the Bitcoin story, the AI story, and then maybe this like more like umbrella approach where you're like, we're just looking for advanced computational opportunities to leverage our infrastructure. Yeah, look, I think from the Bitcoin side, people are attracted to the way we've built

the business, which is long-term infrastructure, no cutting corners. We're building multi-decade data centers. Our uptime and efficiency is fantastic. Our execution capability, we don't miss construction milestones. If we say we're going to do something, we pride ourselves on doing it at least on time, if not earlier. So that is something that I receive a lot of positive feedback from investors, along with the transparency of our monthly reporting and the insight into our operations. The vast majority of those investors are thrilled about the emergence of the AI business. They see it as another great macro tailwind for us to tap into. They see the exponential opportunities in that sector. They see how well suited our

facilities are to service in that market. And I wouldn't call it a hedge on Bitcoin because most of our investors, all of our investors are very bullish on Bitcoin. But the alternative strategy just creates this more robust business in their mind. And that lower volatility should lead to a lower cost of capital in the future and the ability to scale both sides of the business even faster. Because if you've got a commodity facing business that fluctuates really volatile day-to-day, that's going to have a very different cost of capital to something like an AI business where you've got customer contracts, more steady state demand. So the ability to put those two together, lower our overall cost of capital, and then hyperscale both sides of the business

is a really appealing part of the strategy for investors. Given cost of capital is so important to the business, how do you all think about in the United States, kind of the Fed interest rates and some of these drivers of cost of capital versus maybe other markets that you've evaluated, whether you should go into or not and cost of capital there or interest rates and central bank decisions? Yeah, look, I think cost of capital globally, a lot does depend on what the US does anyway, right? So I see the benefit of being located in the biggest capital market, the biggest capital pool we listed on the Nasdaq, as you know, three and a half years ago under Iron, with JP Morgan

City can accord us the lead. And I think being active in that market, it wasn't easy IPOing at the top of Bitcoin in November 2021. But really now you're seeing the benefits of having that public market transparency. And we're not talking about just the equity side, the credit interest that's emerging in this late stage of the monetary tightening cycle as people anticipate more liquidity coming back into the system is becoming really exciting. And I think when you look at our underlying business and the ability to finance a business that is exposed to both Bitcoin and AI, we're getting a lot more traction in the credit markets on terms that may be a lot more appealing for us to scale up this business. And ultimately, that's how you can grow a business

successfully at a lower cost of capital is when you start layering in different parts of the capital structure. How much of the business in terms of building the infrastructure, the cost of capital, the decision between Bitcoin and AI are directional bets on these two trends continuing versus if another trend popped up, you could easily kind of switch to whatever was kind of the opportunity. How much is, okay, the company is going to succeed or fail based on whether bitcoin and ai is you know bigger or five years from now yeah it's kind of both well to be honest like we first bought bitcoin over 10 years ago and that sounds impressive but i bought on the run up to a thousand dollars it went to 500 and i thought this is nonsense and sold it all um but over time

you learn that this thing doesn't die there's only 21 million you can't stop it how does this not create this positive flywheel effect particularly with the halvings every four years where it goes up a lot in value and the penny dropped around 2017 in that regard and for us it just almost feels inevitable the future of bitcoin and if you want to keep it simple as a gold analog and it's far better at being gold than gold is it's scarcer easy to transfer easier to divide all the stuff you know gold parity for bitcoin is 750 000 dollars per coin we're trading at less than 10 the market cap of gold. This is like going back 10 years and looking at Netflix versus Blockbuster and saying, oh, no one's ever going to stop going into a video store and renting

their videos. They're not going to use this online thing. The whole world, the thematic over the last decade has been the disruption of the physical world by these digital social networks. So very simply in 2017, we went down that rabbit hole and just emerged and said, how does this thing not succeed? And how does our next generation ever going to own gold compared to Bitcoin? AI is clearly a very big map for 10 to 20 years. I think it's going to be very profound in terms of how it does change things. The trajectory to get there, not sure. But the bottom line is we're very bullish on both, but we've built this base layer, this bedrock of high-performance data centers that can do any high-performance compute. At the end of the day,

We're experts that manage that physical infrastructure, power density, ventilation aspect, as well as the technology layer on top of that to monetize it. One of the aspects we haven't talked about is sourcing the hardware, both Bitcoin mining equipment, but also the GPUs. I've heard that it's very hard to find the GPUs at times. Bitcoin miners at times, especially in topsoil markets, becomes very difficult as well. How do you all think about sourcing hardware? Absolutely. And I remember when we bought our first NVIDIA GPUs, it's almost like a reverse interview with NVIDIA, where you have to get on the phone and justify why our money's not good enough in isolation and why we should be entitled to get in the allocation of the chips.

And it wasn't easy coming out as someone who hadn't done it before, but we got our our allocation we waited we received them now that our ai cloud service is installed with raving customer reviews to be frank that's gotten back to nvidia we're getting allocations on a far more quick basis so um absolutely the chip scarcity narrative is real um they will try and scale up and what's at the heart of all this pomp is this real world versus digital world dislocation, where I think people don't appreciate all the time the fact that you've got these revenue line and demand drivers that are linked to the digital world, which is exponential. The growth in internet stuff, the ability to grow vertically as you

get adoption of these new technologies, but the ability to service that from a physical world, whether it's the silicon, whether it's the wafers, whether it's the data centers, the mobilization of capital, three to five year grid connections just to get power capacity and build out transformers. All that stuff in the real world has a massive lag and linear trajectory. So in some ways, our whole business thesis has been around monetizing that lag between the linear real world and the exponential demand drivers on the revenue side. What has been the biggest surprise to you in building the business so far? You had the infrastructure background, but what surprised you?

I think it's been the different way so many other people are trying to do things. And that's the beauty of capital markets. At the end of the day, you get all shapes and sizes, all these different business strategies. But what surprised us is no one else is really doing the same thing. No one else has taken that holistic kind of macro view over a multi-decade period and said there is value in owning the greek connections and the power access there is value owning the real estate and the land portfolio there is value in taking the time to develop these specialized power dense data centers that don't have the constraint of traditional data centers but also

have more uh uptime more quality than perhaps a ccan or a shipping container of you know a a Bitcoin mining industry, and that kind of ability to zoom out and then know that you've got this valuable asset base to monetize over decades has been something that we haven't really seen through other businesses. So we've got the crypto miners, you've then got these AI cloud service providers, but most of those providers don't own their data centers. They're just acting as middlemen where they'll sign a customer contract, and then they'll come to people like us and say, have you got data center capacity to install it? And inevitably, you can see how the customer service then fails because you've got an AI cloud customer who all of a sudden lodges

a ticket and says, there's an issue with my GPUs. An AI cloud operator who doesn't control their data center, all they can do is lodge a ticket with that data center and wait for them to come back. So being vertically integrated, controlling the whole value stack is something we haven't really seen anyone else do. And it surprised me. Yeah, it is kind of surprising. I guess people just are scared of or it's difficult to do the capital investment to actually build out the uh infrastructure um my last question is around the energy sources how many of the customers care that the energy is renewable versus they care that it's cheap like cheap renewable energy is always a kind of a phrase that gets thrown around and cheap and renewable go together a lot of times

but do customers care about the renewable component or do they care just about price has more kind of capitalistic uh incentives um it's actually less about the so it's important with bitcoin mining where it's a more competitive industry um but on the ai side the cost is probably less important um because we're selling capacity at you know two to three dollars a gpu hour the cost of power is like six cents a gpu hour so you're on 97 gross margins so whether power is 50% higher or lower, it kind of doesn't shift the dial for us and therefore the end customer. But the fact that we've gone down this sustainability path since day one, where it's not just 100% renewables, which we've done since day one, but it is targeting that excess using

electricity that other industries cannot or are not using by virtue of where it's located, has been a story that's been very appealing to most customers, investors alike, because we're genuinely not placing additional burdens on infrastructure and energy generation and taking that power away from other people. Yeah, it makes complete sense that you've essentially found an edge, right? You found energy that other people can't or won't use. You've got competitive pricing. You've got a lot of optionality in terms of the customers that you serve. And those are usually ingredients for a great business. Dan, I learned from you today, which I always appreciate, where can we send people to find you on the internet or find out more about

Iron? Yeah, so my Twitter handle, at Dan Roberts 0101. Iron is also on Twitter, LinkedIn. We're a public company, not too hard to find us. Welcome any inbounds. Awesome. Well, I appreciate it very much. We'll definitely do it again in the future. Look forward to it. Thanks, mate.