Strategic Mining: Embracing Energy Volatility as the Competitive Advantage

Good morning and welcome to the Energy Stage. Today we're going to be talking about embracing energy volatility as the competitive advantage of Bitcoin mining. I'm John Paul Baric, founder and CEO of MiningStore, and I'll be moderating this discussion.

​

David Chernis, director of flexible compute platforms. I've been living very close to the mining sector since 2018, working with CPower and doing demand response. We quickly realized that miners are the best flexible assets the grid has. They're basically digital batteries that can come offline in minutes to seconds, really pioneering the way for what we're about to see with AI factory design as well.

​

My name is Nick Hansen, CEO of Luxor. We build all sorts of software-enabled services for miners, one of those being something we call Intelligent Miner, which activates this volatility and grabs volatility from the grid.

​

I don't think it's the lights. I think your star keeps getting brighter every time you step out here, Nick Hansen. Scooter Womack, CEO and founder of Vega Energy Advisors. We are primarily focused in Texas and have been in the space for eight or nine years now, seeing the evolution of compute happening in energy-only markets and capacity markets. We've got a lot of great stuff to talk about here, so let's embrace some volatility.

​

Our goal with this first question is to understand the energy markets and break it down for someone who's just a consumer buying power at their house. For most people who don't know, energy prices change and are traded every hour. As a Bitcoin miner or as a flexible load, we're able to trade that energy and participate in those markets. What does energy volatility actually mean? When we say the price of power changes by the hour, by location, and by market condition, what is really happening and what is driving those changes?

​

Energy is supply and demand in real time. The more constrained the supply becomes, either at the transmission node, which is very locational, or at a wider ISO scale, the more price moves. Supply and demand is based on usage, and then you have constraints based on extreme weather, infrastructure, local transmission constraints, and peak usage. Air conditioning globally is the largest load that creates constraint. Below that you have electrification loads, data centers, mining, and AI loads. Certainly in the U.S., data center loads, mining loads, and AI loads are going to be the big ones.

​

Price, whether it's energy or really anything else, is a reflection of risk. It is basically a manifestation of what we call net demand: the ability for generation to be online, the ability for transmission networks to work, all the way down to zonal or nodal distribution levels. It is a combination of many things happening in real time. Any trader will tell you that without volatility, you don't make any money. It is constantly traded back and forth. It is a manifestation of buyer and seller demand and generation. Pretty much every market has evolved that way, whether it's an energy-only market like ERCOT in Texas, or a capacity market like SPP, PJM, or other ISOs and system operators around the country.

​

Energy volatility is actually quite intuitive. At four or five in the afternoon, everybody gets home, turns on the air conditioner, starts charging their Tesla, and all of those things drive up demand. Of course prices go up. Most people don't know that over the last several months, in the early morning hours, power has been negative in parts of Texas. You are actually getting paid to use energy. At Luxor, we monitor that with Intelligent Miner. We tell our customers, hey, you're getting paid for every megawatt hour you use right now, so you should crank up your machines as high as they'll safely go, because you're getting paid to use energy.

​

That happens because in the early morning hours, wind is going, gas generation is on, but there is really nobody using energy at a consumer level. People have not woken up yet and turned on the lights and all the things that start using energy. That produces a lot of volatility. You can go from multiple hundreds of dollars per megawatt hour in the afternoon to negative pricing in the early morning hours. Miners that are really dynamic are able to capture a lot of value from that by turning off during high energy price hours and ramping up during negative hours.

​

There are day-ahead markets and real-time markets. Some miners are bidding into both and are trading those markets with their power. The nice thing about mining loads is that they are running and adding baseload demand to the grid, and when the grid needs it, we shut off completely. We reduce our load by 95% to 99%, and then that power goes back to the grid when consumers need it. We're not contributing to peak power, which is the most expensive energy hour of the day or in that market, and peak power is usually going to be the least green power. That's when you're turning on natural gas peaker plants.

​

Because of miners, Berkshire Hathaway did not need to build over $18 billion of peaker plants because we had two gigawatts of mining load come in and act as an alternative method of reducing load. That's one of the amazing benefits of miners. Why are miners uniquely set up to respond to this versus a consumer at their house, an electric vehicle, or a factory that makes aluminum? Why were miners the perfect ingredients to make the power markets work better and keep prices lower for consumers?

​

At a high level, ASICs don't have feelings. We may not want to have our house at 80 or 85 degrees just because somebody else needs power. The ASICs don't get to talk back.

​

I'm not very price sensitive when it's 90 degrees inside, or when it's 10 degrees outside and my house is 40 degrees. I'm going to pay whatever it costs. Bitcoin miners are purely price sensitive. If it becomes unprofitable marginally to mine at a particular point in time, you just turn off. They don't care. Bitcoin miners do not have feelings.

​

With AI, maybe those machines might have feelings and maybe they don't want to get turned off. That's a whole other story. But a Bitcoin miner does not.

​

Something really interesting happened in a new program we have in my home ISO, which is New York. New York has some very low power pricing around Buffalo and Niagara. We have this new model designed for battery storage called the DER participation model. You have to be very fast acting, and miners are fully automated through partners like Foreman, Jeremy, Hiveon, and others. We're able to dispatch in seconds. We can even follow a six-second NYISO basepoint signal, so we could do regulation with ASICs there if we wanted to.

​

What happened in January is that our miners who were not yet in this program, which offers access to ancillary, day-ahead, and real-time markets for highly automated flexible loads much like a battery, shut down in January and February because power costs were too high. Our miners in this program not only made a lot of money getting offline 35% of the month of January, their net blended cost of power plus DER revenues went negative overall, negative $55 per megawatt hour, to run 65% of January in the most expensive month of the year. It literally inverted the economics of mining because of access to day-ahead, real-time, and ancillary markets.

​

Let's jump into the January freeze. We had a very cold weather event in the Midwest in January. It affected markets in MISO, PJM, and others. During that time period at MiningStore, we were curtailed for 13 days. That curtailment event let us enjoy a negatively priced power bill the following month because of the amount of revenue we generated by turning off. When these curtailment events happen, we are being paid like a generator for reducing our load on a day-ahead basis. How were some of your clients able to capture additional revenue, and in this low hash price environment, why is it so important for miners to be flexible and be in these programs to compete and have lower energy prices?

​

It's different by market. That's a really important distinction. In MISO or PJM, or in a capacity market, loads get paid or act as a generator effectively. If you're not participating as a load and you're not creating demand, then you're directly giving that generation back. It's one-to-one, and you're getting paid the proceeds of that.

​

In ERCOT or an energy-only market, you have to go out and take either a physical or financial position in the day-ahead market, or as a term trader or term position. That's how you get paid back there. In ERCOT, what we experienced was not selling back the physical power, but arbitraging the day-ahead and selling forward because it was so overdone.

​

It gets trickier. It does not get easier as markets and technology evolve. There is more of a disconnect between legacy forecasting systems and everything else. ERCOT, as the grid reliability operator, missed the forecast on one of those days by 10 gigawatts of peak load. That is insane. That's an 18% miss. As a reliability operator, it highlighted the disconnect we have with this advanced disruptive technology versus the legacy systems we have. The opportunity for us was to advise clients to sell into that market, still be able to run, and collect the spread. That is how it manifested itself in Texas. We did not have the extended outages that you had.

​

For context, the entire Bitcoin network, every miner in the world, is probably about 20 gigawatts. So a miss of 10 gigawatts is like half of the entire Bitcoin network. I'll speak about how that actually affected the network as opposed to how it affected miners in the power markets. We saw really big difficulty decreases during that time. People were wondering what was going on and whether miners were turning off because they were unprofitable. It actually had to do with power prices and forced curtailments, which is one of the first times I think we've seen that before. We've seen negative difficulty adjustments in the past, typically because of poor economics, like right now or the summer of 2020. Those were really low hash price environments. But in this case, it was specifically power prices in the U.S. driving Bitcoin network difficulty, which is pretty unprecedented.

​

We saw block times at 11 to 12 minutes and difficulty dropped by around 14%. It was a huge difficulty drop, one of the largest we've ever seen, and to Nick's point, completely driven by weather at the end of the day.

​

You actually have to monitor weather to participate in these things. It's pretty crazy how much data you have to process to be really dynamic. You have to take in all the power prices and all the grid metrics, but then you also have to take in weather data. It is a really complex market. Going back to how volatility is a competitive advantage, if everything were the same all the time, there's really no way to make money. You're not improving your models or getting better faster. Here we are in a very volatile market, and miners are really well positioned to take advantage of that.

​

With new cloud capabilities, these third parties are bringing astonishing forecasting clarity and future-forward visibility into energy markets and predictive weather beyond what grid operators have. We're at a pretty amazing time from a tech perspective.

​

How does the AI pivot change the energy strategy for miners? Most of the public mining companies are now retiring their fleets and building out AI data centers. During that transition, how can AI compute still play in the energy markets?

​

I love hybrid computing. Mining and AI loads are really sympathetic to each other. Miners are years ahead of the real estate investment trusts, colocation developers, and hyperscaler hosts of the world because miners understand energy in a really granular way. It is their business model. Continuing to mine while you build out some GPU hosting is something we're seeing now. I advocate holding on to your infrastructure. If you need to add some AI load to it, great.

​

In my opinion, miners are going to be the innovators on the AI side of the house with direct liquid cooling, immersion, understanding energy in a complex way. The mining community is bringing all of this to the center of AI factory design, while the traditional CPU air-conditioned colocation model is also moving toward these innovations. For miners adding AI infrastructure to their powered land, which puts them three to five years ahead of anyone looking for powered land now, you're already connected. Let us help you design flexibility on the AI side through energy storage. We've already demonstrated that we can flex GPU clusters without pausing the workload and while maintaining the SLAs, with friends at Emerald AI, Mercury, Neural, and our partnership with Bentos and Super Micro at the hardware level. We want to work with you to design flexibility on the AI side of your house as well.

​

ASICs don't have feelings. GPUs actually kind of do. They have a strong opinion about staying on. The reason is training workloads. You can't really preempt those very easily. Maybe there is new technology that allows you to preempt a big training run, but if you're running training that takes a month, it doesn't really matter what the power price is during that time. The output of that work is so much more valuable than the energy.

​

We were having a discussion backstage about what the breakeven price for turning off a GPU would be right now. It's something like $5,000 or $6,000 per megawatt hour. For context, the power price in Texas right now is probably around $50 or $60 per megawatt hour. You're talking about roughly a 100x margin on those GPUs, so you're pretty much always going to keep them on. That's not going to stay that way forever. As we build out more supply on the AI side, it becomes more dynamic. Then the HPC guys need to start learning from the miners how to do curtailments and how to turn down or underclock GPUs effectively.

​

For context, we're used to talking about megawatts, megawatt hours, and gigawatts. On the mining side, it's also kilowatts and cents per kilowatt hour. When Nick talks about $50 or $60 a megawatt hour, that's about five or six cents per kilowatt hour, versus the breakeven for these training models where large HPC and AI guys are talking about $5 or $6 per kilowatt hour, which is astounding.

​

There are opportunities and challenges in there. On a financial level and a messaging level, yes, GPUs have feelings and don't want to turn off. But what do you have to do to be a good steward of the grid and the market you are in in a different way? Maybe if you can't give energy back to the grid, you're constrained in that way. The opportunity, as we bridge through margin compression with these workloads and buildouts, is that we need more hedging and more forward term commitments from these folks. They're going to be running these models, so at that point you are incentivizing grid-level generation and capital investment for the ratepayer and for the people in our homes. You're not becoming a burden for all of those costs. It doesn't matter what the real-time price of power is if you've already taken a forward position in power. That could be $50 or $60 a megawatt hour for years, and you just run. You don't have to worry about the economics or stewardship of giving back power. You express your stewardship of the grid in a different way. That is a message that needs to be picked up with these workloads, especially as we face new boogeyman conversations around AI and data centers in communities across the country.

​

No compute will ever be harmed in the flexing of this AI factory inference load. It is always going to run and maintain the SLAs. We would never stop a training or inference workload. These are small incremental curtailments, just a little bit, like a hash rate optimization example. That's the perfect way to frame it. In the hybrid model, all the flexibility is on the mining side of the house.

​

As public miners and private miners move their Bitcoin miners to AI, how do you see Bitcoin mining and AI loads working together, understanding that there are commitments miners have made to the grid, and there is a transition period and different infrastructure upgrades needed? Even when they are running AI loads, utilization of inference loads is not always 100%, but sometimes the power contracts require closer to 100%. How do you view this transition and utilization specifically of energy?

​

Taking it from a trading and risk perspective, you're going to need a hybrid strategy. You can't just go binary, where you're all this way or all that way. It doesn't take too much sophistication, but you have to truly understand the use case, then plan accordingly, model accordingly, and contract accordingly. From the volatility opportunity we see, it is in term trading for these loads. We advise and give access to all of our industrial clients about where forward marks are in different ISOs, where you can hedge power, buy power, and sell power. It's really dynamic. You can buy power for multiple years, sell months, sell a week, sell specific days, and get in and out of positions. I think these loads are going to become active forward-term traders.

​

Miners, this is kind of unironically true, I believe right now might be one of the best times to buy mining in the last few years. Every AI machine being plugged in, B100, B200, B300, is most likely taking a Bitcoin miner off a rack in a lot of places. Companies doing the AI transition are pulling machines off to plug in AI machines. That means there is less competition for Bitcoin blocks than there has been, relatively, based on how much demand there is for power for other types of workloads. They're scrambling to plug in every AI machine anywhere they possibly can, and those are pulling mining machines offline. It does feel like we've reached a bottom point. Any machines that have been taken off are already taken off. Right now may be a really good time to be a miner, even though it really hurts right now. But that's the best time to buy.

​

When you're in mining and you're running your returns, and Bitcoin was $120,000, the model looks amazing on paper. Today the returns look horrible, but that's assuming the output stays the same, difficulty doesn't keep going down, and the Bitcoin price doesn't go up. As a miner, you have a long perspective on Bitcoin. You're hoping hash price goes up and recovers. The previous three years resulted in hash price crashing because many public players were incentivized to simply grow megawatts at no cost. Now those players are going to be removing themselves from the market effectively.

​

One quick point: all of the big miners you see, CleanSpark, Marathon, Iris, were formed during the summer of 2020, which was one of the worst times to be a miner in memory. Now look at them. It could be the time. Maybe I'm talking my own book here.

​

I think it's a great time to start mining. I'm thinking about starting mining. If you understand energy right now, with the depressed cost of ASICs, it's a great time. I'll take another angle from a regulatory perspective. We have Senate Bill 6 in Texas, which mandates that large loads over 75 megawatts have to be flexible. We're starting to see other regulation in other areas. If you're a miner, you already have a great relationship with the grid operator at the ISO level and likely at the utility level. If you're running hybrid load, both mining and AI, they are more likely to give you a capacity increase because you've got all that flexibility on the ASIC side. That's really valuable to the utility and to the ISO. It's a great bargaining chip and a great way to partner with the grid operator and the utility. We love to see hybrid mining and a good part of mining load stay there as some AI infrastructure comes in.

​

Last question: is energy volatility in the United States going to increase or decrease over the next three years?

​

I'm bullish on volatility long term. It is a requirement of where we are with all of the loads coming on and with the types of generation we have available. Acute volatility will show its head again. Sustained long-term volatility for multiple hours, maybe not outside of winter and extreme events. But I'm bullish on volatility across the country.

​

Everybody is trying to lower volatility, but they're also adding tons and tons of generation and load. I don't actually know. It's a great question to think about.

​

I think more and more volatility. What we're trying to do with demand response as the near-term solution is get from today to 2030, when more generation comes online. We're still losing generation in the U.S., nuclear and fossil, and we don't have a bridge to get to 2030 right now. Small modular nuclear and micro are way off. I think there will be more volatility and more need for flexible resources, and with that, greater and greater value for highly nimble, fast-acting demand response resources like mining and AI compute.

​

Thanks again for listening in, and remember to mine on.

​

Thank you.