Join host JP as he speaks with Carson Smith, an early Bitcoin miner and engineer, pioneered immersion cooling with fish tanks and later spearheaded SBI Group's global mining arm, building out over 250 megawatts of capacity, where he developed internal monitoring tools and implemented demand response programs for grid stability. Currently leading 21 Tree Capital, Density AI, and Merkel Edge, he focuses on ultra-low-cost energy for Bitcoin mining and anticipates power costs becoming equally critical for AI data centers due to rapid chip depreciation and market commoditization. Smith emphasizes robust treasury management to navigate bear markets, stressing the importance of cash reserves, and offers entrepreneurial advice on learning from inevitable, even multi-million dollar, mistakes and persevering.
Johnpaul Baric interviews Carson Smith, pioneer of immersion cooling and builder of SBI Group’s 250MW global mining arm, now driving energy-efficient Bitcoin and AI infrastructure with 21 Tree Capital, Density AI, and Merkel Edge. They unpack power economics, demand response, treasury discipline, and the entrepreneurial grit required to survive and learn from multi-million-dollar mistakes.
00:01:00 – Early Mining Journey & Technical Evolution
00:03:00 – ASIC Development, Fried Cat, and DIY Immersion Cooling
00:07:00 – Scale of Early Immersion Operations & Challenges
00:10:00 – Career Pivot to SBI & Building a Global Mining Footprint
00:17:00 – Large-Scale Deployment Lessons & Mistakes
00:20:00 – Operational Tooling for Massive Miner Fleets
00:24:00 – Negotiating Energy Contracts & Local Partnerships
00:26:00 – Demand Response & Grid Dynamics
00:32:00 – Creditworthiness & Capitalization Challenges
00:33:00 – Current Focus: 21 Tree Capital, AI Crossover, and Merkel Edge
00:36:00 – Power Costs & Strategic Importance (Bitcoin vs. AI)
00:41:00 – Small AI Data Centers, Offtake, and Idle Capacity Monetization
00:46:00 – Converting Bitcoin Mines to AI & Compatibility Challenges
00:47:00 – Cooling Trends in AI Infrastructure
00:49:00 – Treasury Management & Preparing for Cycles
00:52:00 – Entrepreneurship Wisdom & Closing Thoughts
Enjoyed the episode?
If this conversation brought you new insights, we’d love for you to leave a review, it helps more Bitcoin enthusiasts and professionals discover the show.
Have a guest you'd like to hear from or a topic you want us to explore? Drop your suggestions in the comments. We’re building this podcast with and for the mining community.
@blocksmither
[00:00:00] Jp: Carson Smith is an early Bitcoin miner who went from an engineer in a tier three in tier four data centers for telecom and high frequency trading firms to co-founding and running SBI groups previously. SoftBank Investments, multi ex hash mining arm before launching 21 Tree Capital,. A single family office building high efficiency Bitcoin and AI data centers, and investing solely in Bitcoin native ventures and infrastructure and co-founding density AI slash Merkel Edge, a data center development and energy company offering the lowest price energy for data centers in the us. He prices power in Satoshis pioneers immersion cooling in stranded power sites, and treats Bitcoin 21 million supply cap as the era's most reliable monetary constant. Carson, welcome to the show.
Carson: Oh, thank you. Glad to be here.
Jp: So tell me about your experience in these data centers and , did you learn about Bitcoin while you were working in them and when did you start realizing that mining [00:01:00] was for you?
Carson: Uh, I realized mining was for me early on. , I've always kind of been a nerd and a tinker at heart. That even started, well, early on when I was in elementary and middle school. , I came across mining, , sometime during my college days. , I was playing with, , SETI at home. And someone on the forms mentioned, using idle, , computing power for mining this, , digital currency of the future. , Study at home was using idle computing power to do protein folding, to find some type of medicine , or maybe a cure for cancer or other things , in biology. I wasn't much of a biology guy. I was always more engineering computer science. , that side, but it was always an interesting concept, and that's what originally brought me into Bitcoin.
, At that time there were no data centers. , It was just people running at home or like me in their college dorms, which is what I did., I expanded a little bit. I started, I tried at that time , in CPUs, but by that time it was already had moved on to GPUs. Did a little bit of GPUs, for a while.
, That [00:02:00] worked and then people started moving to FPGAs. I tried to design an FPGA, but I was too slow, , for that. Then next round, quickly came Asics. I got some of the first generations of Asics. Some of them, I got, some of them I was burned by, lost a lot of, Bitcoin. And those are extremely high numbers
and, today's value. I try not to look back too much, but, , hindsight is always 2020.
Jp: And so let's talk about this asic development of yours. So was it with fried cat? , , How did you get into building your own,, circuit boards or chips? Of where did you sit in the stack and, , were you successful in getting it operational?
Carson: let's first focus on development and then also mention Fred Catt. 'cause that's, there's another interesting story there as well. , So on some of the development, a lot of that really never really got off the ground. So at that time I was studying electrical and computer engineering. At NC State University and I worked at some of the classes and stuff, I worked with semiconductor physics and design and mastering, embedded electronics. , I did some designs with FPGAs and I was in [00:03:00] college at the time, so I was trying to use what I was learning. that I had just learned maybe like weeks or months before, that is a long time ago. So I can't quite remember when exactly it was now. But , utilizing what I had just learned to like try to build an FPGA because, , I knew I could, make some optimizations if I could just get certain number of things right. , But the design and my limited experience and actually building. A big FEGA. , I wasn't able to get that off the ground in time By the time, , that Asics, , started hitting the market and that goes into fried cat. Fried cat was actually, , I ended up buying a lot of, Asics from Himm, the asic miner, the blades. There's these blades. I still have, , dozens and dozens of them. , Back in, North Carolina, which is where I'm from. I still have those. And I ran these, blades in, , fish tank aquariums that I bought either from, well, I bought from both Lowe's and from Walmart. I got shelves from Lowe's. , I used an aquarium fish [00:04:00] pump that, I can't remember if it was Lowe's or Walmart. , The pump. I used the tubing, but then I needed a radiator. None of them sell the right radiators. 'cause you gotta actually remove the heat from the oil. So for that, I went to frozen cpu.com. I found the biggest radio I could and I made it connect to the Fish tank Aquarium. Tubing and then I use kind of fans like actually computer fans inside the fluid along with the fish tank pump to circulate all the oil that had dumped into these aquariums to circulate it through the radiator. And then I just lined up all these boards and a fish tank, , aquarium. And I had, several shells of them. This was running well, and, but these asics were also very expensive. And the, of course, just even like many years later, the hash rate was shooting up, especially as asics were taking over and replacing GPUs. I spent a lot of Bitcoin, , both Bitcoin and dollars, , into buying, new Asics. Some of these asics were like butterfly [00:05:00] labs, KNC minor, and ended up spending a lot of money into those, and a lot of those were unfortunately never delivered. , ASIC design is very expensive. Especially when you get to like just the risk wafer phase that a lot of the r and d, that's the really expensive part. Once you get past that and you generate a good asic, then it's a lot cheaper to produce, but they have to charge a higher price because they have to make back all their initial r and d costs and of course all the costs for their employees, maybe some of their marketing and all the other overhead. But once you get to that phase, the actual chip cost is cheaper. That's why some chips you can buy really cheap. But what these companies did is they didn't have the right teams and such to fund this. So they funded it all through pre-sales. And I think probably the story, what happened is some of them maybe ran away with the money, some they tried to design and it was an ultimate flop.
, I really don't know the full backstory behind all of those. And, , for some of that it makes sense where some miners were delivered, some not. , This can go where you have like a wafer, [00:06:00] you get yield problems where a wafer, some parts of the wafer will produce viable chips and some parts will not.
And so if they were expecting like a certain amount of production from the wafers that they had ordered based on the money that they had coming into their company. Then the yield was much worse. Then they don't have enough to, supply the orders that they promised. , I'm not saying this is necessarily the case, but this could potentially align with some of the cases that like, , some people saw in Isol where some minors were delivered and some were not. This was like 2013, especially mid to late 2013, early 2014. Is also around the time of Gox. , , There was a lot going on at this time.
Jp: That's right. When I was getting into this space, 'cause I actually bought a butterfly lab, jalapeno miner. It was a five
Carson: I still have one of those.
Jp: So you got that, did you buy it? , And that was the two week problem, for them it was always like, two week delivery, we'll be delivering in two weeks, as , the biggest kind of like, , scam going , on the message boards , and the memes back then.
So , you got into mining, I mean the fish tank [00:07:00] operations. Sounds like you were immersion cooling. So is it clear to say or you're one of the first immersion cooling minor, if not the first.
Carson: I think I ended up talking to someone, , at some point who tried immersion cooling before I did. I can't remember that clearly. , So then they may have been the first immersion, or if I'm remembering wrong, then maybe I am the first immersion cooler. But could definitely say with confidence, I'm one of the first, so I experimented with different types of oils.
, I just bought like oils online and I think I was just buying from Amazon. , At the time and I just bought like large jugs of 'em and just filled , the tank, , with oil in it. And it probably like wasn't the best at the time, but I was working a lot using some of the engineering knowledge. I had a lot of, , trail and error and experimentation.
, And just going from there, went to using oil because the air cooling wasn't quite working 'cause I was working in a smaller, limited space, especially when I was either at a dorm room or at home. And I wanted to [00:08:00] expand a little bit , and I had some access to some warehouse space as well.
My family is, runs a logistics company, so I was able to utilize a little bit of space there, but most of it was in smaller room settings. , The heat, the cooling capabilities, as well as the noise. , And properly, like just aligning everything on all the shelves. Everything was just became a lot easier.
When I did immersion cooling,
Jp: How many devices are we talking about? It's hard to measure in Gigahash or mega hash back then, but how many devices were you trying to run at a time?
Carson: it was several dozen.
for the asic miners that actually got from fried cat, which I never talked to 'em later. I have no idea what happened to 'em. So fried cat, if you ever see this. Hello? , I bought a lot of miners from you at the time. But, , so buying those, I ran several dozen of those.
, A few others that I did. , Some of the GPUs were radiant. They were the. Seven, nine seventies or seven, nine nineties if I remember correctly. I might be wrong on that. Don't hold me to it. , But that's, I think [00:09:00] that's what I was running, , when I was do doing GPUs. And then I was running several dozen, of these miners and , at first I was running most of 'em on air cold setups.
And then I moved to the immersion setup. Then I also went to going from home and having gigantic vats of oil and then realizing also what a pain it was to dispose of that. That was actually a problem that I didn't think I had on, at least at the time,
but I was very young and naive Then.
Jp: most people don't think about that. So you're going from one of the first immersion coal miners ever working with Fried Cat, who those listeners who don't know, , sold a lot of these first ASIC chips and collected tons of Bitcoin in pre-orders for a batch , and basically ran away, kind of moved off the assist off the blockchain.
Disappeared. But you, Carson, you've stayed around. Where did you go after this?, And how do you end up at SBI, which is a huge achievement in launching with
Carson: Yes. So in some ways [00:10:00] I actually didn't fully stay around and that's how I ended up there. , Because it wasn't an industry at the time, it wasn't a career like you, you couldn't fully make a career yet. Out of it. I mean, , you could make money, but even then it was still difficult because just like it is today, you have to properly maintain like your cost of power, your cooling and everything.
And at that point, , I was very new to this and that gave me the proper exposure and learning and applying that to, with the engineering concepts , of just who I am. And, but also too what I learned in my schooling, oh, and then the concepts that I learned later on in my career.
So now, how did I end up to where I was in my career out from going from mining? So 2012, 2013 was around the time that I graduated from NC State University. While the time I was at NC State, I did study abroad in Japan. I really liked living there. , Many good things about it. Good food, good safety, a lot of things that drew me to wanna go back to live there.
And, but around the same time though, rather than doing for asics, because the, as I just mentioned, [00:11:00] a lot of the. Original asics, , manufacturers or the people who tried to manufacture asics ended up flopping and I gave them money. Lost a lot of money. I was very discouraged. I almost lost a lot of money in Mount Gox.
I missed it by like just a week or two. I had some delays of trying to do they wanted some extra KYC documentation. This was just like one or two weeks before they went belly up. And I kind of saw this , and during this process too, I had also started going down the Bitcoin orange rabbit hole and learning about the history of money economics.
And I mean, I have a slight autistic streak as well, so that too also. Powered me further down the rabbit hole to learn a lot. And so I really believed in the ethos of Bitcoin and what it could become later. But I didn't wanna scare myself and it wasn't a career yet, and it couldn't pay the bills that I needed to pay, especially with, , trouble with Mount Gox happening, a bear market now in suing, trouble with delivering an [00:12:00] Asics.
, , And I wanted to go to Japan, . All that together drove me to go to Japan and start a career. The first company I worked with was a telecom company. They were an ISP and a cable TV provider. they provided, , their ISP services. They would do fiber up to a point, and then for the last mile they would use copper, coax seal cabling.
And I managed the servers, , mostly DNS and DHCP servers, , for that ISP, in their tier three, tier four data centers. , Primarily tier three. And then after that, there were several reasons for that company that, I wanted to move on and look for something better. And so through a recruiter, I found S-B-I-S-B-I used to be SoftBank.
Probably the SoftBank. Everybody knows the SoftBank Vision Fund. One of the CFOs, one of the finance guys at SoftBank previously, many, many years ago, was handling the investments and that's created S-B-I-S-B-I was originally stood for SoftBank Investments that eventually spun off into its own company.
It went public. [00:13:00] Then SoftBank sold its shares , in the exchange, and they kind of divested and became mostly two separate companies and SBI group. Now, what was originally SoftBank investments, but now known as SBI group, became a FinTech conglomerate, almost kind of like a fidelity maybe in some ways, like a Morgan Stanley.
They do like asset management,, high frequency trading exchange. . They own and lease cargo ships and airplanes, insurance, all kinds of stuff. There's like 270 to 300 Cary companies now. , And in securities exchange, they're the largest online securities broker. So I ended up working for them. I worked for their high frequency trading exchange as an engineer.
I have a strong background in Linux. I can do a little bit of programming, not enough as like a full software developer, but enough to make me slightly dangerous. And utilizing that to, basically tweak and develop code as infrastructure and work with the coders and the traders doing the high frequency trading algorithms [00:14:00] and the fixed trading income desk and at various points within the company.
I worked within, those groups, , as an engineer. At some point, they became interested in Bitcoin as the market started moving up in 2016, 2017. They knew about my previous background that we just talked about in Bitcoin, and they asked me for help. So I gave them some advice and helped them in setting up, teaching them what a wallet was at that time.
, All the Japanese guys who are executives, they're in late forties, late fifties. They have no idea what a wallet is or what coins or Bitcoin or any other cryptocurrency is. , And they're wanting to set this up, and capture this, new change, new revolution in finance. Some of the younger guys knew, but they still didn't quite grasp it, right?
I had developed already a good relationship with several of the, executive and SVV team. Of the group companies. And through that and them knowing my background, they asked me for advice to set up, teach them what a wallet was. And now of course they expanded and they're well versed.
They have a [00:15:00] good infrastructure for that. But in the very beginning, right, I had to set essentially like an advisory and setting up the exchange and then telling them, okay, who's the right people they need to hire to do this all properly and secure? During this process, , I started, I was talking, I developed a, as I said, a good relationship with the chairman and the SVPs.
I pitched internally along with some others to start a Bitcoin mining arm, and we were successful and getting them to approve that, and they ended up investing a few hundred million dollars into doing that and building out more than 250 megawatts around the world. That ended up, , building out in Sweden, in Iceland, in Virginia, in Texas, for example, Texas was originally, it was in Rockdale, what later became the site that now is operated by Riot.
Also did some sites in Kyrgyzstan, in the Middle East, in Russia, and just all over the world for a [00:16:00] while. , In that process, after starting it up I started as operations manager and then I eventually, , was promoted to CEO and I did that for a few years and that basically kind of sums up my most of my career at SBI , and how I kind of moved to that position from my early Bitcoin mining days to Bitcoin mining again.
Jp: The 250 megawatts. I mean, that's , some huge scales. Well, if you can you dive into that, just the, launching a new site, you're, , running the bitcoin mining operations across multiple continents. What are some of the struggles that Bitcoin miners of that scale face, and maybe specifically where was the industry in terms of large scale operations?
, How big was Genesis?, Was Mar just getting into the space? Kind of where were you SBI compared to other competitors? 'cause when we talk about 250 megawatts, that's obviously a huge amount
Carson: No. So that built over time, , a lot of that was built, between 2017 and 2021. , And then of course, in those early days, 250 megawatts, it still is a massive amount, [00:17:00] but it was also a much bigger amount than it is today. I remember one of the first, entities that we looked at. We actually , to quickly jumpstart.
We looked at acquiring existing entities. , For example, one of the entities that we went and looked at was a company, called Project Spokane, running out of Missoula, Montana. And there I met a few other guys, one who many in the mining industry would know, Kevin Zing. Who is the SVP at Foundry?
. He was one of the ones behind that site. , And he later moved on. He had some disagreements with management and he moved on to work at Greenwich and then Foundry. , There were a few others. Some of the guys like at Hash House were also connected to that. And then we looked at other sites at least as starting out as in acquiring at that, going back to that site.
I think at that time, this was October, 2017, they were running about 20 megawatts and they built this on a very low, almost like a shoestring budget. It was very interesting site to see. . Speaking of Genesis two, we ended up, , investing in a company. When I say we, I mean [00:18:00] SBI group. It was done through another SBI, , group company into, a data center company out of Europe who, , we partnered with to build data centers in, , Iceland.
And they ended up also. That process acquiring a very small Icelandic entity. And we built a data center within like a stone's throw of Genesis, , data center just outside of Keflavik, or actually in Keflavik, just outside of re, , revic. And we built, in northern Iceland. So it definitely expanded a lot , and in that, , time, 250 megawatts was definitely a lot larger, than it is today.
And through that process, , we learned several mistakes along the way. I mean, , we made those mistakes. We learned from them. I learned from them. And, I'm still
like taking what I learned and, and,
Jp: What are some mistakes that come to mind?
Carson: , So some of the mistakes too and just how, , some are contractual, , some of the engineering design, some, and being careful of like how the power, infrastructure is done.
I [00:19:00] mean there are several , that just, it's there. There's a lot , when you go through this size and you go through several procedures. Some are large mistakes, some are small. And no matter how small it is, even for successful operation, there is always. Some, , thing that can go wrong and something that will go wrong no matter how successful operation is.
And each one of those points turns into a learning opportunity, and that's what we were able to capitalize on, and that's definitely what I've taken to heart and continued to take into the future.
Jp: And , as you mentioned, every infrastructure project is always not on time. It's really hard to get them on time, especially if it's your first one of that scale.
Carson: Yes.
Jp: We're talking about managing tens of thousands of units. So did you guys develop internal tools to do that? Were you relying on third party tools?
And what operational knowledge maybe that didn't exist before in the fish tank existed in the 250 megawatts, of deployments? How did you get from, , the fish tank to that?
Carson: So at that [00:20:00] time, maybe foreman and some of those other, if it's okay to mention specific names, but maybe some of them, , existed at the time, but I don't recall them. A lot of it was like, , for each of these data centers, , they built their own. Local systems to manage. And in this case, using my programming and my engineering, my Linux background, I have some Python experience, bash experience, I have networking experience.
, And that's through my studies in high school, through my studies in university as well as my own hobbies. I partially wrote the system. To get one of the first data centers started that, that we got running and the first one that we started, we ended up not buying an existing one.
We ended up, \ building it. And what I ended up doing was using something like, , Very cheap, but smart switches cheap, so we could save on cost, but smart switches, so we had access to remotely manage the switches and see the Mac tables. We had a system where each port went to a specific physical place and I wrote a [00:21:00] script.
That could scan the entire network. So even though each machine was DHCP, it had a random IP address, so we couldn't necessarily tell what it is without making a blinking light, which we could do. You could do a blinking light, but when you have 20,000 machines, that takes a long time to find the blinking light.
Especially when they're actually all blinking in one way or another. These are all s nines at this time. The s nines they glow green and they have some blinking, right? So finding the right blinking light, is not always the best method, , or especially when we need to locate a specific device.
, And so what we did is we created this method, , , I wrote a script. It would occasionally just run as a batch,, or what's called a CR job. From a local server and it would scan the network, pull the Mac tables, see what machine, what IP address was connected to what port, because we physically designed each switch from a, like port one goes to this physical location on the rack like.
This rack, the first position is port number one. The next one's port number two, the next one's port number [00:22:00] three, and we go down that shelf. And then the next one in line is like port, it goes port number eight, and we go in specific order. And then the switch names were named after the rack. So if you know the switch name and you know the IP address, you can instantly find it within this list of tables.
You know exactly where the minor is. And , we did that and then. Building on top of that where we, the miners had an API exposed, we would give information about like hash rates or temperature. , We built that, plugged it into like open source systems like Grafana and, I don't think it was Prometheus at the time.
I can't remember what, , system backend we use for Grafana. We use something else and we use Grafana to graph it. And we basically put this up on large, , television screens in the, knock or in the monitoring room. And we use that to like monitor the data center both locally and remotely. And this was like built from like a hodgepodge of like different scripts that I wrote and some of the other,
[00:23:00] engineers and software developers wrote, and then eventually, right? , You have more unified solutions like a creator that was bought by brains you have for and a few others. then there are a few other companies that are now developing some that, , I probably shouldn't say, but, , yeah, so
Jp: space has come a long way.
Carson: It's definitely, it's come a long way from what it was before where everything, when we built , the data centers , back then, like early stages when I was doing the aquarium tanks to even then in 2017 when we were doing, , 15, 20 megawatt data centers, it was still like not to the level where it is today.
, And we were kind of like having to like basically hack together a bunch of scripts to make it work.
Jp: It makes sense. And obviously it's the progression of the industries, the maturity that you've been able to see throughout your time, as a miner. What about negotiating some of these sites and leases and contracts, any, is there, what can you talk about back then , on how that process worked and maybe give any details on.
The difference between [00:24:00] negotiating a Bitcoin mining lease or an energy contract as a Bitcoin miner versus a traditional data center, or what some of the things you learned for people that are considering negotiating their own energy contracts that are still applicable today.
Carson: A lot of that too. , We worked, , and we empowered, , and definitely, worked very closely. We found local partners to work with. And the reason we did that is local partners are well connected in the community . And that community , can include like the utility companies as well as, , people that you would need to like hire jobs.
, So when you're working with government, a lot of, like, for example, government workers, like improving zoning rights or some permit or regulation, what they like to see is they like to see jobs. This is like a big thing. Another two that we saw, and we saw even early on when we were looking in 2017 at the project in Missoula is we also have to be careful of, even though a zoning condition may be right, we have to be careful of like sound issues.
, And this one, like, it was actually an industrial zone and there wasn't really much residential. There [00:25:00] were a few residential plots in industrial zone. That's a kind of a thing in of itself. But they also had some other issues where acoustics would bounce off the mountains and you would not hear it in one place, but here in another place, miles down the road.
, But then without, even without those acoustics, you have to think, for example. Of how it may affect the local, , population. So we have to be careful of things like that. , For power. , When we're doing grid contracts, it's very similar to what, , a lot of companies do today. And that, it forms like mining typically forms as a base load, for power to help secure a stable load of power.
As , a grid provider, a grid maintainer job is to maintain supply and demand to power. Both are important, otherwise you get blackouts or brownouts. If you have too much power but not enough demand, you have problems and vice versa. And so Bitcoin mining, runs 24 7, but it can respond with the software on it.
It can respond [00:26:00] within seconds to changes in supply for power infrastructure. That allows, , grid providers while working closely with them to help them balance the grid. So this is a useful application for grid providers. So, many of them are now,
Jp: like that, demand response, implementing demand response with SBI, you guys were one of the first people doing it at enough of a scale where grid's like you actually make an impact locally here, running 10 20 megawatts. So how do those conversations go in the beginning, in the early
Carson: In the first stages, they didn't, I mean, they had some understanding, but they fully didn't understand it because they didn't see,, there's not often many, like large industrial users that can like, just scale down, like Bitcoin mining, for example. Other large industrial users that, like some Bitcoin miners used to run out of and some that we actually repurposed was like an aluminum smelter.
An aluminum smelter cannot scale down. If you scale down, the aluminum's gonna solidify and it's gonna ruin your entire factory, pretty much. Or at least that's my understanding of it. I may be wrong, but that's my understanding. [00:27:00] But , in other cases, other factories, you have a factory line. , You can't just easily scale down where Bitcoin mining, it's just a simple command to the firmware on the devices and you lower the clock frequency of the machines or you, or in some cases you cut off power to the machines and you can.
Basically instantly scale down to meet the power demand, such that, like if there is a power plant that goes offline and there's not enough power for the grid, the bitcoin miner that is taking power from the grid can then scale down their operations. So now there's enough power for residential homes and such to, , so that grid is balanced.
And this helps provide a financial incentive because anyone coming in to build extra power for a grid. That, whether that's state money, private money, , even if a state money, there needs to be some type of return on investment. And when you have different, , variable factors to consider and what the cost of power may be, it can be hard to financially model.
To an investor what their return may be. [00:28:00] Or with Bitcoin mining, you can model it as a, for a power contract, you can model it as a very flat rate that the power contract will return. And this provides a very stable modeling for a power utility provider, building a plant to model on and they can model a return on investment and structure , that in a right model that fits for their investor profiles.
, And we worked , in educating. Grid providers. This was difficult in some ways early on. , We also worked with a company called Lanum. This was a company that SVA group invested into. They're based outta Texas. , And Lanum started, , in educating, , ERCOT about this program., Lanum two. We also, did this.
Utilizing, wind farms, we were looking at wind farms in West Texas. , If you look at kind of wind maps of the United States, where most wind blows and is produced, that area, kind of west Texas, I think it's like New Mexico. That rough area of the US produces a lot of land, but [00:29:00] not many people live out there.
So you get a high supply of electricity, but a low demand problem. Now you need transmission to transmit all the electricity to like Dallas or Houston or other major population centers. But those transmission lines weren't well built out and wind farms kept building because they got subsidies from the government.
So they would further oversupply power. This would create a problem because now this is a problem for the power grid on maintainers in balancing that supply and demand. So sometimes the energy prices in West Texas would become negative due to these, , wind forms, and sometimes they would, , adjust their fan blades.
So the, that they would kind of resist the wind and not flow. And sometimes they would let the. Farms generate electricity anyway, and they were so profitable because they were receiving subsidies from the government. We would come in behind the meter and offer, say like, Hey, we won't pay negative prices.
We'll actually pay a low price, but a base load so that you don't have to sell at a negative [00:30:00] price. And then, we'll actually, in some cases, if we negotiate the term, if the contract. We would negotiate the contract such that if the price becomes a reasonable amount that you can sell to the grid, to factories, industry homes , and other users, then you can sell to the grid.
We'll turn off our data center and we would use low cost miners to do that and, we would essentially provide like a base load for that., In doing so, that kind of helped expand some of that, , growth there in west Texas as well as, , educate Ercot and other power grid providers in the US on the benefits of these types of programs.
speaking of too, that company Lanum, , they recently signed, an agreement with Stargate, like, which is also a SoftBank, but the original SoftBank and, , open ai. And they're doing campus for their AI development together with Oracle and Stargate and those guys in, Abilene, Texas.
Jp: And you
see
all these previous crypto miners, some of them did make it to the ai, big leagues, you [00:31:00] know, core, core weaver being one of them. And, as you mentioned, Lanum Cruso. So there's plenty of those groups that are in development with ai. You mentioned one thing about the credit worthiness really, and Bitcoin mining.
To these power plants and to the grid operators. When you were with SBI, how was the credit conversations with these local electric providers, especially state owned electric providers, knowing that Bitcoin was still a little bit on the fringe, how often did that come up and how many of them maybe didn't believe you were gonna be around in a few years?
Carson: That's definitely the case. For us it wasn't as much trouble. But that's definitely the case for the lot of miners. And even today, , for example, if I go try to raise money now with Bitcoin mining, there's enough still firms even today and like private equity firms that won't trust the credit worthiness.
Bitcoin miners, even public Bitcoin miners. Like even if I were to like secure a public Bitcoin mining client, with a data center I'm building. And I try to use like some firms as well to help, come into the [00:32:00] cap table to raise money. There are some that are still distrusting of that. And of course some pub co miners, I won't say names, have economics that they need to improve on, , and their operating costs.
But, in terms of capitalization. Especially for smaller miners. It is a concern. And often what will happen is, , there'll be large deposits, especially for PPAs, for collateral, and this will require significant cash reserves, at least on credit. And we still had to do this at least on credit worthiness.
We, behind us, , we had SVI group, which is a financial conglomerate of 200 7300 companies. Right. . It's publicly traded. It's audited by Deloitte. , It is a public company, so everything's all public, so we didn't quite have that problem. But it is definitely, when you're negotiating those kind of contracts and credit worthiness becomes an issue, it can be a and can be a problem.
And , there are several creative ways to do it, and that can differ depending on state, depending on grid provider, [00:33:00] your power broker. That can be things of depositing money for collateral, using some of your assets, as collateral if you're able to do that.
And there's different methods to do it. , But that is still an ongoing issue today.
Jp: So let's transition to , where you are today, Carson, and you're at 21 Tree Capital and any of your other endeavors and projects. Tell me more about that , and what do you see for the next, , couple months for you guys and building Bitcoin mining? Are you still actively involved and , what's on the top of mind for you today?
Carson: So I'm still actively involved in Bitcoin mining. , I would say I'm 99% Bitcoin mining. And the reason I say I'm 99% is Bitcoin mining. , Is in my blood, , and is what I want to do. I mentioned there's, we talked some of ai and some of that comes to, in that, I'm not looking for necessarily for a turn play to ai.
But, , some of the investors that I'm talking to want to do some AI and at the same time as Bitcoin mining. So, , utilizing some of the techniques that we learned in operating at [00:34:00] low cost, some we cannot replicate for ai. , , some very specific differences and that goes into another long conversation that's too long for here.
But, , there are some specific differences, but there are also some techniques that we can apply that we've learned over the years from Bitcoin mining to AI build outs and management. And that helps with, , what some of those investors want, in terms of building up new data centers. , So 21 Tree Capital is really just basically a family office investing , and primarily in either, , infrastructure, Bitcoin.
And mining infrastructure or Bitcoin companies, some of the Bitcoin companies that invest to, like even public companies like Meta Planet, it has investments into those. , Going into that below that, some of the companies as we mentioned before, are like density ai. Density AI does AI and , Bitcoin mining and then Merkel Edge.
Merkel Edge is a US based company, with , my partner who's a longtime oil and gas, working at several, , large investment bank and private equity [00:35:00] firms doing hundreds of billions in, , gas, oil and gas deals and asset management. And through that,,. We're looking at some renewable energy, but also focusing on a lot of oil and gas and utilizing stranded gas where we're able to, achieve cost well below 2 cents, in some cases 1 cent , and maybe even below.
, And we're doing this at scales of larger than 10 gigawatts. However, we're doing this too, in that. So we have exclusivity in that we're the sole developer. We utilize our DA data center, , experience to develop data centers to do that, and then to purchase gas to develop data centers. However, we're not able to develop 10 gigawatts of data centers ourselves.
, That's a lot of money to go in just for ourselves. So we're also to like talking to others to raise money, to further build out data centers, some, and even that full capacity. We may not be able to bring everyone on the same cap table, especially miners who are competitors to each other. So we're talking about two, breaking out in different areas, , into like different projects [00:36:00] or different legal entities where it's like, okay, one miner here, one miner here that we're working with and or like an AI or data center company over here. And that's the same case. Two, like maybe a data center company doesn't wanna work with miners. But going back to the heart of it,, our focus is on, bitcoin mining and building out the infrastructure for, , the next, , or actually not the next, but the current revolution and, , digital finance.
Jp: So you haven't left and you're still building strong. And it sounds like to your point, you have , big visions and very cheap power costs. How important is it, obviously, the power costs and the bitcoin mining industry, AI doesn't matter as much, but what are you seeing in this transition and , how important does it matter?
Carson: I think AI will matter, and some of the AI companies I talk to are forward looking and they think the same thing. Maybe I'm wrong and maybe they're wrong, but in some ways Bitcoin mining in the early days didn't care as much about power costs as they do now. Ai, I think they will as well. , Several months ago, Jensen, CEO of Nvidia, he [00:37:00] came on at a, I think he was at a conference and he called himself the Chief Revenue Destroyer Officer.
He's made comments referring to like how he wants to have like, generations of chips that come out like every year or every two years, every three years. Sounds familiar, right? Sounds like miners. So you'll have AI companies that will have to reinvest, , substantial CapEx into buying new AI chips every year.
With AI chips, when you think of offtake clients who wanna use that, they'll wanna use the latest and greatest AI chips. You'll still find some , who don't mind older ones, but most of your clients will wanna use the newest technology. So the oldest technology, in order for them to use that, you'll have to lower the pricing, which lowers your revenue on it, and that severely, , affects the, , margins, , for running older ships as well as to the depreciation, because now, your assets are depreciating fast, you need a quick return on investment and all these kind of dynamics together along with a growing increase.
And, , the number of competitors in the [00:38:00] market space will just further kind of commoditize the market and in some way that , commodity works , and how this, , kind of, This whole structure works, right? Is that I think it will push, the incentive for lower and lower power prices down for all or most AI companies.
Maybe I say most, and that some AI companies still may be relevant, that they don't fully care about as much power. They may care about, very specific like locations to population centers that may be their like niche and what's always important to them. But I think over time, even the ones who aren't caring about power will care about power and they'll care about it more and more.
, As I just mentioned, , some that I'm talking to now are already starting to see that and look forward and seeing that and from mining. We already do that now. And we've done that for the last several years, and that's a key driver in any, miners metric whenever they're planning a new build.
And if you're looking at this and you're planning [00:39:00] a new mining build one of your biggest, factors that you should consider is your cost of power. And you should consider that cost of power too. Not just your pure cost of power and how much money you're making each month, but also consider your cost of power and consider that too against your comp.
In some ways, your competitors in mining, they're both your competitors and they're also collaborators. Everybody works together. To strengthen Bitcoin , and that produces, that helps support Bitcoin and drives Bitcoin price higher. But you're also competing for the same, , Bitcoin rewards each day. And part of that too, with a main driver, your cost being the cost of power, you need to consider your cost of power in relation to, , the cost of power for other competitors.
So for ones, if you're paying, , like 10 cent power, 11 cent power for mining. I doubt anyone is, but that , you should look for other uses. Things like that. And I won't go into specifics too of like what you should use, because sometimes there's different dynamics of like maybe, , of [00:40:00] just certain , where you utilize some power for mining at some point of the day and some points of the day use that power for other like factory other resources.
And this kind of complicates the dynamics of your business model. And so every business model is different or some. You may use the heat, like district heating, use the heat from miners to heat local homes or, hot springs or things like that. , That can change, the influence that , for power costs , and o other, , factors into your business model.
Jp: so Carson, I think you hit on, a few great points there. The first one being that the, just like miners move from site to site over time and from lower efficiency miners moving, from higher cost of energy to lower cost of energy as their life cycle goes through, GPUs are going to do the same thing.
, And as we scale out, the applications and the customers , for AI is changing and is, , moving, I guess. , Up closer to smaller businesses, away from your cloud providers. So always still be there, but basically more customers are [00:41:00] always gonna come on the AI train and trying to bring their own AI resources and owned services.
Needed to their customer. So my question is regarding the size and scale of these developments. Obviously we're seeing massive gigawatt data centers across the board. Do you think there is going to be a need or for five megawatt, 10 megawatt, 20 megawatt sites for ai or is that unique to Bitcoin and can you talk more about that, maybe why those edge locations, where they exist and who their customers are?
Carson: So going for ai, some of the AI companies that I've talked to, , that have done like very specific like direct requests for offtake that we may potentially build for. like I said, we're still mostly building for Bitcoin mining. That's what we really care about. But, , we're still talking to some ai, as we just to maximize the power resources and data center development capacity that we have and our expertise.
But going back to talking to some of the AI companies, some of them are. , Looking at to do direct offtake agreements [00:42:00] for smaller amounts. And those smaller amounts have been like five megawatts. , There's one looking at five. There's another one at 15 and there's another one at 20. And then if you look at to like the full cost for developing an AI data center, , if you're looking at like.
Two. When I say full cost, it also depends on like from what voltage , and what, , levels of responsibility that you're including. But, kind of a general rule of thumb, like if you're looking at NVIDIA chips, you're looking at something of like 25 to $30 million per megawatt. So it's a very high cost for that.
, like some of those direct off takers. , They're looking at, smaller sizes and going back to like five megawatts. And then there are a growing number of like cloud providers or like middlemen that kind of.
Function like an AWS, but for ai, , there is actually some use, there is some, parts of AWS that you can use for GPU computation, to do like AI based workloads. And then there's also some other, , companies that we're trying to partner and work with, what they do is they help other small businesses that may need a very specific [00:43:00] AI model for whatever they're doing.
Or maybe it's not an AI model, , maybe it's just some type of like. Not ai, but some type of rendering, physics rendering or rendering for motion capture or video content creation or something of that nature. I mean, there's tons of ideas, tons of things to do. , And these, , providers work almost work like a middle arrow, AWS and that they route, , a request for computation to data centers free that they have, , collaborated and brought into their network.
. For a data center, like if you're building like a one megawatt, two megawatt data center that's AI focused, this allows you to connect in this platform and either fully utilize that capacity as much as possible. It may be hard to get a hundred percent capacity and it generally is, but you can try to get as much capacity as possible.
Or you could also to go to other companies who want to secure an offtake agreement where they want to use your AI data center. For the GPUs that you have for some compute capacity, , or some physics [00:44:00] rendering, if it's not AI for or for whatever. And then you may have some idle downtime where they're not utilizing your capacity.
And what you can do is you can actually connect this data center to these providers such that it'll utilize the out the idle capacity and you can set priorities such that your contract for primary offtake, they will always get priority. Then if there's ever a downtime, where they're not taking, , compute power then, and there's a request coming in from the cloud provider, then that compute power will go to the cloud provider and the cloud provider essentially, , as acting like a middleman, takes a cut in the middle and they route to data centers around the world.
India, Europe, us. They do filters, like if you need, , filters for privacy restrictions that some European companies , , may care about or have to follow, they route only their, compute loads to European data centers. , You can do different filters like that. And so there is a growing market for that as well, and that allows both, , small and [00:45:00] large data centers to, work together.
Jp: Thanks for that explanation, and , I definitely agree with you. It's so expensive to build one megawatt of Nvidia chips. There has to be , a market for that type of compute, and that compute is going to be adequate, , for a large number of applications, even though it's not a hundred megawatts. Of capacity.
, My one question regarding, , the difference between or upgrading existing mines into ai, we saw Core Scientific do that and pay about $8 million a megawatt to upgrade their infrastructure. Is that feasible for most miners or is it, , put it down back to the pad , and start at a
Carson: I would say it's generally not feasible. I mean, there are some portions, especially some core electrical infrastructure that you can reuse, but a lot of like the racks and everything, you will have to just end up , tear down, and redesign. , There are a few companies and we are also working on. , Some designs as well and trying to make some designs that could be [00:46:00] reusable for both.
And even then, , like if you make it fully reusable for both, like with very little change that's, , that's,, something good that , we hope to achieve, or at least something in the middle where, there is a design that, , is capable of supporting minors. Also too, it can support AI without like major overhaul of the data center.
That would cause like a large amount of downtime to change and a large amount of costs. But this can also get complicated because you may have different form factors for what, , an AI server may provide, may use at one time, and then the different form factors of all the, , bitcoin mining companies and.
Of them, as , they all like to change different form factors or really push their own, , without naming any names. And sometimes , they change and they use, , slightly odd, factors. Like even when doing like traditional rack design, some like to do something slightly different than what the rest of the, , market does.
And what this does is [00:47:00] it makes it hard to design, like cross compatibility. But then there's also whole number of other issues too, where, it can make it difficult to just very easily convert, , an existing Bitcoin mine to a data center, for ai.
Jp: And do you see with AI, most people moving towards liquid cooling or water cooling or potentially immersion? I haven't seen much about immersion. How does cooling play an impact in all of this?
Carson: That I don't see yet that I still see a lot of air cooling. I do see some trends and some people looking at immersion. , But I see no like large trend there yet. But I think we may, see a growth in immersion cooling. I have seen like a few immersion cooling designs grow and we're trying to look and, see if there's some that we can do ourselves.
, And I see that's a possible trend that may pick up. It's not something I quite see yet. , But then again, as I said, I'm still focused on 99% of Bitcoin mining, so I also might be missing these trends, just [00:48:00] as, , to throw that out there. And there may be a trend that I'm missing by focus on Bitcoin mining.
There may be some of those trends in AI that like towards immersion, , that I am missing out on. , And so, but
Jp: And that's totally reasonable. You can't do everything.
Carson: do notice some of that. Yeah, , you can't do everything.
Jp: And being in the space for so long, Carson, how do you now manage running your own business? The volatility, , payroll, power, expenses, CapEx, especially during Bitcoin drawdowns. Talk to me more about treasury management and what you've seen as best practices slash what you're implementing today, , in your organization.
Carson: , For one, for the core of the treasury and the core of everything, fo as I said, focused on Bitcoin mining. Volatility is vitality. , Who was it? Maybe, I think it was Dylan Le Claire who coined that phrase. Maybe it was sailor, was someone who coined that phrase, but I really like that. , And that is at the core, but we also do, , look carefully and we make forward looking projections several years, and we consider bear markets, and maintaining a, proper amount of cash.
would [00:49:00] like to hold everything in Bitcoin, but we still have bills denominated in dollars, and some of those bills are like power bills. , Some are utility bills, some are payroll, some are taxes, and those are all $10 million and we need to be ready to pay those bills and we need to be ready to pay them in bull market times and in bear market times.
And as we know, , I've lived through many market cycles. A bear market cycle will always come and it'll come again this time. There are many people, I think, saying that a bear market cycles won't come anymore. I hope so, but. I doubt it. , The history for the last few years of Bitcoin has said otherwise history for the rest of markets and the world history for hundreds of years have said otherwise.
And so I'll continue to follow as if another bear market is coming and looking at, Bitcoin. Like if you're planning Bitcoin bear markets, right? If you look, they typically go from big drops from the all time high. Like , where the all time high is like. If we've reached there yet, or , if we're going to reach there later.
, I don't know. , I expect and hope that we're not at the all time high yet, and that's gonna come, [00:50:00] sometime later this year. But from the all time high, it typically drops between like 75, 80% every, , just about every single time in every four year cycle. So being able to handle that drop is important and being able to make sure you can pay your bills even when it drops that 75, 80% and pay your bills, not get scared and not have to sell your Bitcoin, , is important.
, That because those times are very likely to come again. But also one thing to throw out is that doesn't also necessarily mean it's gonna drop 75% from now because I don't know, like now, if we're at the all time high, if the all time high is a million dollars and it drops 75%, or if the all time high is $110,000 and drops 75%, those, you end up at very different numbers.
But. The point ,, is to maintain like a significant, and detailed plan of what kind of cash payments you need and having a cash reserve to be able to make those payments for at least several years in the future. And planning that based [00:51:00] on at least , a four year cycle, should provide you a runway to keep going forward.
Jp: Thanks all that advice. Carson, I think as anyone listening to this, can see your expertise, obviously from mining as a college student at NC State where we both went to school just at two different times and all the way up to s. I , and now the new projects you have working on, , is there anything else you want to talk about on the podcast and mention to the viewers about mining or AI and or entrepreneurship in general?
Carson: , I would just actually go for entrepreneurship in general. You're gonna make mistakes and some of those are gonna be small. Some of those are gonna be big. I made some really big mistakes and by big mistakes, I've made some multimillion dollar mistakes, , and I've learned from those, thankfully.
I hope you never make multimillion dollar mistakes, right? And , you may not too, like also have the ability to make that level, but I was working in a much larger company, right? , Where I had the ability at least to make that, but you'll make a big mistake in a different way. So it may not be a multimillion dollar mistake, but it will be a big mistake to you and it'll make a big impact on your life.[00:52:00]
It'll make a big impact on your business, and that will come no matter , how you plan, something of that nature will come and I can, almost guarantee that it will. And so being prepared for that and then also using that as a learning opportunity and , getting back up and trying again.
Even if you fell completely. Get up and try again and, you'll eventually succeed.
Jp: You will succeed. We all will succeed. And if you ride the wave of Bitcoin. It's easier to succeed than riving the wave of like the dollar or something that is not necessarily growing like internet and ai. So thank you again, Carson for this time. This was amazing to have you on the Digital Gold Podcast.
And remember guys to mine on.