An economic analysis of bitcoin mining when using an intermittent, renewable energy source like solar power.
Written by Daniel Frumkin - any opinions expressed in this piece are his own.
Just 6 months ago, I wrote an article for Bitcoin Magazine titled The Next 10 Years of Bitcoin Mining which described how the mining industry is rapidly evolving as large institutions, energy producers, and governments become increasingly involved.
A lot has happened since then. Uninformed hit pieces in mainstream media publications, controversial Elon tweets, and the 100th iteration of China bans have all shined a spotlight on mining and made once-rare knowledge about the industry’s dynamics now somewhat commonplace for Bitcoiners. Among this knowledge is the possibility for bitcoin to incentive renewable energy development around the world by providing a means to monetize surplus energy which would otherwise be wasted.
While I’m already aware of bitcoin mining being used for curtailment of surplus hydro, geothermal, nuclear, and wind energy, I hadn’t heard of it being done at any noteworthy scale for solar energy. And so, as I saw solar getting mentioned more often lately, my curiosity got the best of me and I decided to do some analysis to determine its viability. In this article, I’ll share my findings along with some commentary on the whole “green” bitcoin topic.
One important point to mention before we get into the financial analysis is that the “clean” and “dirty” descriptions of various energy sources can be misleading without deeper context. For example, manufacturing and deploying solar panels and batteries requires energy-intensive mining for minerals from the earth, using toxic chemicals, and burning significant amounts of energy in the manufacturing process as well. Once manufactured and transported, solar panels can produce clean energy for many years, but there are still noteworthy environmental costs and social costs to be paid which are rarely mentioned in the public discourse.
In fact, one of the biggest production centers for solar panels in the world is Xinjiang, China, which is largely due to the fact that Xinjiang has cheap coal that can be used in the manufacturing process and thus bring down the cost of production. Incidentally, “cheap” labor may be another not-so-convenient factor that’s bringing down that cost of production in Xinjiang, but that’s a whole other topic.
Ironically, Xinjiang is very often singled out by critics of Bitcoin because of that heavy denomination of coal used to power bitcoin miners there during China’s dry season from October - May. (At least, before the latest ban may have put an end to mining in Xianjiang for good.) More generally, however, the most commonly cited metric when looking at bitcoin mining’s environmental impact is the percentage of total network hashrate that’s powered by fossil fuels. This too can be misleading without more context.
For instance, the US Energy Information Administration estimated that the amount of natural gas being flared and vented in the US is around 1.48 billion cubic feet per day (Bcf/d), equivalent to about 1.52 trillion BTUs/day (British Thermal Units). In more common terms, this amount of natural gas could be used to generate roughly 162 TWh/year of electricity. Based on the latest figures from the Cambridge Center for Alternative Finance shown below, waste gas in the USA alone is likely enough to power the entire Bitcoin network, which consumes an estimated ~116 TWh/year. Not to mention the venting and flaring that doesn’t get reported, or the waste gas in other countries around the world.
Combusting natural gas and using the electricity produced to mine bitcoin prevents that gas from being flared or vented, which in turn prevents methane emissions which are estimated by the Environmental Protection Agency to be 25x worse for the environment than CO2 over a 100 year timespan.
All of this is not to say that efforts to transition more bitcoin hashpower to renewable energy sources are pointless or bad. To the contrary, any scenario where bitcoin mining can be used to help build and scale renewable energy production with greater economic efficiency is exciting. Rather, our goal is simply to help everybody approach this topic with a nuanced perspective, understanding that raw energy consumption amounts and even the denomination of that energy which is renewable does not actually tell us everything about the final environmental impact of bitcoin mining.
With that context in mind, let’s shift our focus now to the topic of bitcoin mining being used to help scale solar energy projects.
The Bitcoin Clean Energy Initiative (BCEI) led by Square and ARK Invest recently published a whitepaper which explains how bitcoin mining can be added to solar power + battery systems to help scale them beyond what would be possible if there was no way to monetize the surplus energy produced during peak sunny hours. Since I’m no expert on solar power, I’ll be relying on their data to get realistic inputs for my own mining profitability analysis.
Specifically, there are two points that are extremely relevant.
The paper includes the $/kWh electricity price for solar using the LCOE metric, so setting an electricity price for this hypothetical mining operation is easy. I’ll just take the average from the range, which is $0.035/kWh.
The much bigger challenge is determining how much uptime the mining machines would have given the intermittent nature of solar energy generation and the fact that most of the energy it produces would be consumed by the grid or stored in batteries rather than used for mining.
In typical mining applications, it’s assumed that the ASICs will be running basically 24/7. There are exceptions, such a load balancing programs which sell energy to the grid during times of peak demand, but generally even these will have 80%+ uptime. This is important because there is one big external cost in all of this that could make or break the use case for bitcoin mining: capital expenditure (CAPEX) for purchasing hardware.
ASIC hardware has historically depreciated over longer periods of time as the network difficulty increases (i.e. the BTC mined per terahash of hashrate goes down), so downtime is extremely costly as it pushes out the date to break-even on that initial CAPEX. In a case of too much downtime, it’s possible that bitcoin mining will be a net negative on the balance sheet of an energy project, meaning that it never produces enough profit to pay off the initial investment.
BCEI’s paper links to an open-source model incorporating real-world data that serves as a proof-of-concept for a solar system that integrates bitcoin mining. The model is back testing the use case with historical Bitcoin network data and incorporating it into a much more complex financing scenario which is beyond the scope of this piece.
The purpose of this analysis is to isolate the bitcoin mining portion of the project and see how competitive it would be with more typical mining operations. Instead of back testing with historical data, I will be making forward projections incorporating data from ARK’s model into standard bitcoin mining profitability calculations.
A key excerpt from the the model’s README is the following:
“The logic of the model is optimized to prioritize meeting grid demand. That is, the sun's energy will not be used to mine bitcoin unless the demand from the grid is first met. Once grid demand is met, the model assesses whether it is more profitable to store energy in the battery or mine bitcoin based on trailing profitability levels.”
Basically, the model is making a sophisticated calculation about what to do with surplus energy at any given time. This makes the ASIC uptime calculations very complex as deployed hashrate is fluctuating hour-by-hour, whereas our goal is to get a single figure which we can use to approximate the average hashrate for the operation over the total time period being analyzed.
In order to get a realistic value, I simply took the average of all the hourly values in the Deployed Hashrate column of the model, resulting in a final value of 1.923 EH/s. Since the maximum hashrate for the operation is 5.449 EH/s, this average hashrate estimate equates to 35% uptime for the deployed ASICs at the mining farm. Put another way, calculating the mining revenue for a 1.923 EH/s operation that mines 24/7 will give us more or less the same results as for a 5.449 EH/s operation that mines for about 8 hours per day on average.
Now we have all the information we need to carry out some basic financial projections.
One of the critical components of forward financial projections for bitcoin mining is the network difficulty. This is because the mining revenue generated per unit of hashrate goes down as difficulty goes up, which it has done at a rapid pace as shown below.
Since 2017, network difficulty has increased from 317 billion to 25 trillion, equalling a 8.42% monthly increase over the past 56 months. Considering the current semiconductor chip shortages and unknown future price action, I’ll conservatively set the monthly difficulty increment to just 6% in my calculations to simulate difficulty approximately doubling each year.
Note that I will set $30,000 for monthly OPEX (operating expenses), which is also a very conservative estimate for all the costs involved in staffing a 170MW facility and maintaining the hardware in it.
Below is a 4-year cash flow analysis which uses the CAPEX and ASIC specifications from the open-source model. Power consumption is set to 35% of the maximum consumption in order to maintain the same average W/TH efficiency of the ASICs used in ARK’s calculations.
We find that with difficulty approximately doubling every year while price remains constant, this investment performs poorly with a -99% annual internal rate of return (IRR) if the hardware has almost fully depreciated by the end of the 4 years—and it does not factor in the halving that would occur about 3 years into the analysis. This is our baseline, but it’s not actually very realistic. Difficulty is unlikely to keep going up at the historical pace for 4 years if price doesn’t also perform similarly as it has in the past.
Still, improving ASIC efficiency and today’s large profit margins for miners mean that difficulty is likely to continue going up rapidly for at least the next 1-2 years in all but the most extreme bearish scenarios. In other words, even if BTC price remained constant for 2 years, difficulty would continue increasing until the average cost of production for bitcoin miners equals the actual BTC price.
But what about the bullish scenarios where BTC price also increases rapidly? Well, let's set a 4% monthly price increment and find out.
Now we see that mining remains profitable, albeit at an ever-decreasing rate, for the entire 48-month period (again, halving not factored in). Furthermore, the initial CAPEX investment is actually paid off near the end of the 3rd year of operations. This tells us that adding bitcoin mining to this solar project would be rational only if the investors believe that BTC price is going to increase significantly in the next 4 years.
And since that’s the case, we might as well check what would happen if the miners were to HODL some portion of the BTC they mine, say 50%, rather than cashing it out entirely to fiat each day.
Now we’re talking! By setting a HODL Ratio of 50%, the CAPEX break even point is shortened to less than 2 years and the final cash flow (pink line) of the operation including the value of the ASIC hardware inventory (yellow line) exceeds $400MM. The takeaway is simple: the success of this hypothetical mining operation is highly dependent on BTC price. No surprises there.
But wait, there’s one more caveat here. Given the initial BTC price of $40,000, we should also compare how this hypothetical investment would perform vs. simply buying and holding BTC with the $247,680,000 CAPEX. In the mining scenario, a total of 5153 bitcoins are mined (see Total BTC Mined in the STATS part of the image above) over the first 4 years of operations (not accounting for the halving around the end of Year 3). But at $40,000 per BTC, the $247,680,000 could be used to purchase 6192 bitcoins with no OPEX for simply holding. Phrased differently, the miners never break even on the CAPEX in BTC terms.
This conclusion holds well in different projected scenarios because BTC price and difficulty are correlated to each other. So if difficulty doesn’t increase as quickly as projected, it very likely means price has performed poorly and the investment doesn’t do well in fiat terms. If price does increase substantially, difficulty will likely continue increasing at about the same pace as it has for the past 5 years, in which case the investment doesn’t do well in BTC terms.
One other point to reiterate is that the analysis here is isolating the mining portion of the solar + battery + bitcoin mining project. Although these numbers may not look attractive, there are other factors not taken into account, such as the possibility of government subsidies for renewable energy projects as well as access to extremely cheap capital which can make this look more reasonable to large companies with big balance sheets who are able to tolerate more risk and longer payback periods on investments of this size. Basically, those trends I talked about in The Next 10 Years of Bitcoin Mining are becoming more noticeable.
Anyway, before moving on to the last part of this analysis, I want to drive home further just how much of the risk:reward depends on bitcoin price. So let’s look at one more visualization—this time with FREE electricity and a constant BTC price. One could argue that the surplus energy produced by the solar panels during peak sunny hours would otherwise go to waste and have zero economic value, so we should consider (almost) all mining revenue as profit. In this case, what would we find?
Unfortunately, even with free electricity the operation never breaks even on the initial fiat CAPEX investment unless BTC price goes up substantially.
One more thing worth looking into before we wrap this article up is a comparison of the same hypothetical bitcoin mining operation as above, but with a constant energy supply rather than an intermittent one.
This means that we will use the maximum hashrate and power consumption from the model, 5448960 TH/s (~5.45 EH/s) and ~178 MW respectively, as well as the same CAPEX. However, to make things more interesting, we’ll increase the electricity price of the hypothetical full-uptime operation to $0.05/kWh and keep the $30,000 monthly OPEX.
We can see that even without any BTC price appreciation, the operation breaks even in just over 1 year and remains profitable for about 2.5 years. Assuming the miners were to simply liquidate their hardware inventory and stop mining when they are no longer profitable, they would have a final cash flow of $85MM—about 34% returns on the initial investment. This is a significantly less risky investment than the 35% uptime operation.
And if we add the 4% monthly BTC price appreciation and 50% HODL Ratio into the calculation that we used for the most profitable scenario with the solar project, the upside is… very nice.
CAPEX break even occurs in just 9 months and the final cash flow of the project exceeds $1.5B. The total BTC mined over 4 years is about 14,600, more than double the amount that could have been bought with the initial CAPEX investment amount.
In summary, a mining operation with $0.05/kWh electricity and full uptime drastically outperforms one with $0.035/kWh electricity or even free electricity but 35% uptime.
One other variable not discussed here that could be interesting to play with is the type of ASICs used. In the model from the BCEI paper, a mix of old-generation (Antminer S9), mid-generation (Antminer S17 and Whatsminer M20S), and new-generation (Antminer S19 and Whatsminer M30S) is used. If the goal is to minimize risk, the investors could avoid expensive new-generation miners, sacrificing some efficiency and longevity in order to have a much lower initial CAPEX investment. Perhaps with excellent timing (e.g. purchasing 50,000 Antminer S9s around the halving in May 2020 when they were selling for $20-$40 each), it could work out well.
Furthermore, the miners in the model are only running when their is surplus energy to consume directly from the solar panels. If the miners were to use battery power or a secondary energy source to increase uptime, particularly in the early months of the operation before network difficulty has increased significantly, that could also improve the probability of breaking even on the mining CAPEX.
The table below shows the months to break even on the CAPEX for the mining operation analyzed in this article with a range of electricity prices and ASIC uptime amounts. (Note: NaN means that the CAPEX break even does not occur in the first 48 months analyzed.)
Mining farms with 90%+ uptime are likely to break even within the first 2 years of operations with electricity prices at the high end of the spectrum, while extremely cheap electricity prices are not enough to make most operations viable with less than 60% uptime.
This article may not paint such a rosy picture of bitcoin mining being integrated into solar projects, but all hope is not lost for a relatively green future for bitcoin mining around the globe. Market conditions can change to make this more feasible in the future than it is today, such as a decrease in hardware prices if more manufacturers can become competitive with MicroBT and Bitmain.
Meanwhile, other renewable energy sources such as hydro and geothermal are already a big part of the bitcoin mining landscape, and wind energy is potentially more realistic as well because it can have more consistent generation than solar.
At the moment of writing, hundreds of thousands of ASICs have recently been deployed in the Sichuan and Yunnan provinces of China where they will be consuming surplus hydroelectric power produced by dams which have been overbuilt there. In Russia, Canada, the USA, and potentially elsewhere, miners are consuming increasing quantities of surplus natural gas which would otherwise be costing energy producers money to flare and vent while also emitting harmful methane. Load-balancing programs for urban energy grids are gaining popularity as well, making grids more efficient and robust so that they are capable of handling periods of peak demand.
The energy consumption of the Bitcoin network may be trending up, but that ultimately does not tell us much about its actual environmental impact. Even talking about the denomination of hashrate powered by renewable energy fails to account for green use cases like consuming waste gas, or reusing the low-grade heat output from ASICs for other residential and industrial applications. This is all very complex and nuanced, but one thing is clear: we need proof of work to have a meaningfully decentralized global monetary network. To those who don’t see the value of bitcoin, I encourage you to dig deeper.
*Additional note: The average efficiency of the ASICs used in these calculations is 32.6 W/TH. This is most similar to an Antminer S19 or Whatsminer M30S+, which are the current newest-generation of mining hardware. Based on recent market prices for these hardware models of $80-100/TH ($ per terahash), the CAPEX for hardware in the analysis done above would be $400MM-$550MM, as opposed to $248MM (the figure used).
Bitcoin mining software company: Slush Pool, Braiins OS+ & Stratum V2.
By miners, for miners.
Increase hashrate on S9s to 17+ TH/s, improve efficiency as much as 30%, and get 50% lower pool fees on Slush Pool
Reduce data transmission between your farm and pool by 95%. Configure parallel usage of multiple pools. Set a backup pool for the whole farm.
It focuses on making data transfers more efficient, reducing physical infrastructure requirements for mining operations, and increasing security
It focuses on making data transfers more efficient, reducing physical infrastructure requirements for mining operations, and increasing security