On The King's Trail
TVA Supremacy & Mullet Mining
Below is a conversation between myself and Tom Masiero, co-founder of Kungsleden, a digital infrastructure company recently acquired by Cathedra (TSX-V: $CBIT; OTCQB: $CBTTF), a public bitcoin miner, where he is now a member of the board.
Maverick: Please give Second Power readers a short overview of your background.
Tom: I’ll start off my story at the point at which I was introduced to Bitcoin. In 2016 I was working as a software engineer at a digital advertising firm. A guy named Roger Ver solicited our help for a few of his companies. This sent me on a wild intellectual and professional ride that led me and others to start a bitcoin mining venture called Great American Mining (GAM). GAM’s strategy was to mine using stranded and flared natural gas resources. A couple years later we sold GAM to Crusoe Energy Systems which was one of the first bitcoin miners to pivot to powering AI related compute.
After GAM, a partner and I founded Kungsleden, a compute-center—data center is inaccurate and lame—hosting and development company for bitcoin miners and AI related compute. We’ve since been acquired by Cathedra, a public, micro-cap bitcoin miner.
Maverick: Your twitter bio says you’re an “ERCOT disrespectoooor.” Why does ERCOT suck?
Tom: It’s a bit of a troll to my Texas friends. In truth, I have a lot of respect for ERCOT but its been oversold as a great place to mine. Some miners have and will continue to succeed but there are many mining grave yards littered throughout the state. It gets really hot in Texas which shrinks the useful life of your ASIC fleet. To slow down the decay, operators need to invest a lot of capital in cooling systems, cutting away at already tight profit margins.
Plus, the lucrative demand response programs that have been a big attractor for miners, are a direct consequence of bad energy policy. The wild variability in available load in ERCOT is the predictable end-state of building gigawatts of unreliable solar and wind power. Miners are a blessing for ERCOT, because of their flexibility, but the variability of these sources make ERCOT’s power generation extremely volatile and the system very complex and difficult to participate in.
I prefer the TVA.
Maverick: I’m guessing the TVA is superior because it has more reliable generation?
Tom: Exactly. Unreliable solar and wind power make up less than 5% of TVA’s energy mix where as ERCOT is much more dependent on these sources. Nuclear, natural gas, coal, and hydro are all more reliable and play a much larger role in TVA’s power generation. Like all systems there is some variability, so the TVA isn’t immune from that but the volatility is less severe and operators are afforded much longer lead times to prepare and respond to drops in output.
In ERCOT, you’re expected to respond within seconds. Miners can do this and will only get better at it but its without a doubt more stressful and costly to run systems that are that responsive. Although, I will admit that it acts as a strong forcing function for miners to develop extremely flexible and responsive loads. So that would be a positive externality of this pressure. But, from a pure business perspective, the climate, reliability, simplicity, and longer lead times of the TVA is superior.
Maverick: The concept of mullet-mining or mullet-computing (AI in the front, Bitcoin in the back) has recently gained more familiarity and popularity. Can you add any nuance to this phenomenon?
Tom: Hyper-scalers like Amazon, Google, Microsoft, etc operating general LLMs and working towards AGI aren’t going to be mining bitcoin anytime soon. They’re in their own hyper-competitive battles and have no interest or incentive to mine bitcoin. Plus they’re only going to gobble up compute centers that have Tier 3+ levels of redundancy. This is overkill for a bitcoin mining operation. The GPUs powering AI applications that consumers are familiar with such as ChatGPT, Claude, etc will not be inferencing next to hashing ASICs anytime soon.
So when we’re discussing mullet-mining or mullet-computing we’re talking about bitcoin miners and hosting providers expanding to more commercial and narrowly defined AI related compute. A simple example of this would be an AI model being trained on just oil and gas related information that OpenAI wouldn’t have access to. The loads for these species of AI are more flexible and don’t require the level of redundancy that their hyper-scaling, generalizing brothers demand.
Maverick: Thanks for your time Tom. Please share any contact or social media information for those who want to follow you and/or Cathedra.
Tom: No problem, and thanks for having me. I’m @FarmahTom on X and Cathedra’s handle on X is @CathedraBitcoin.