September 19, 2020

Analysis of the network hashrate and power consumption of Bitcoin mining

Mining is a fundamental component of Bitcoin as a network and as an asset. But for all its importance, mining remains one of the least transparent and understandableto the general public of parts of the Bitcoin ecosystem. The purpose of this report from BitOoda is to make the composition of Bitcoin miners more transparent, ultimately helping to better understand the health and health of the system.

Bitcoin mining is a mysterious industry about whichthere is very little information publicly available. Even more or less sophisticated crypto investors often find gaps in their understanding of mining and the potential investment opportunities of this industry. Despite brilliant research from Coinmetrics, Coinshares, and the Cambridge Center for Alternative Finance, some questions remain unanswered. The BitOoda study described here was commissioned by the Fidelity Center for Applied Technology to complement and develop previous studies and try to find answers to new questions. The full report can be found here (English, PDF).

Part 1: Analyzing Bitcoin's Power Capacities: How Much, Where, and At What Price

In this first section, we will try to measure,establish the location and price of the energy capacities of miners, as well as assess their profitability. Having conducted over 60 conversations with miners, equipment manufacturers and sellers, as well as based on data from 45 open sources, we tried to provide as complete a picture as possible of what power capacities are available for Bitcoin mining, where they are located and how much miners pay for the consumed electricity. ...

We then investigated the question of howmining capacity may grow in the future depending on the availability of electricity, the efficiency of mining rigs, and the constraints that may be imposed by the price of bitcoin, the availability of capital, and semiconductor chip technology.

We estimate the bitcoin mining industry has access to at least 9.6 GW of electricity.

Our assessment is based on the following logic: On May 10, just before the halving, Bitcoin's hash rate reached 136,098 petahesh / sec. and on May 17 fell to 81,659 petahesh / sec. We admit that, to some extent, these extreme values ​​could depend on mere chance - so a number of successfully and quickly found blocks could artificially overestimate the calculated hash rate, just as a decrease in the speed of finding a block can also partly be a consequence of chance. However, we eliminate this random factor from our model and make simplifying assumptions to get a rough idea of ​​how much power the Bitcoin network is using. We are assuming here that the source of all the hashrate left after the May 17 lows comes from the more lucrative, new generation of S17 hardware, which includes Bitmain's Antminer S17 and T17, Whatsminer M20, and devices from Canaan, Innosilicon, Ebang, and others. We also assume that all devices unplugged between May 10 and 17 were older generations of less profitable S9 class equipment (like the Antminer S9 and Whatsminer M3).

I draw your attention to what we usethe designations “S17 class”, “S9 class” and “S19 class” as categories that include both devices from Bitmain and equipment from competing manufacturers with similar characteristics. We refer to the class names through Bitmain models because they were the dominant player in the “S9 class” and, to a lesser extent, “S17 class” hardware market. In addition, in all relevant calculations, we proceed from a PUE (Power Usage Effectiveness) value of 1.12, which means that for every 1 MW spent directly on Bitcoin mining, another 120 kW is spent on maintaining the entire associated infrastructure, including cooling systems, lighting, servers, switches, etc.

ANDBitOoda used classification of equipment by key classes.
and: BitOoda, Bitmain, Canaan, MicroBT, Halong, GMO, AsicMinerValue.com.

The figure below shows that if allthe hashing power that operated on the network on May 17 belonged to the S17 class, then they would consume 3.9 GW of electricity. Also, if all 54 exaches / sec. The hash rates that were turned off between May 10 and 17 belonged to the older generation S9 equipment, which would correspond to an additional 5.7 GW of electricity. We make these simplifying assumptions in order to arrive at some general understanding of the industry, while realizing that in reality, although most, but not all of the disconnected powers belonged to the S9 class, and some small part of the remaining capacity is probably S9 equipment. operating in markets with very cheap electricity. The key factor behind the capacity cuts was the drop in profitability of equipment as a result of halving, coupled with the timing of moving mining equipment from north to south of China to take advantage of cheaper electricity (see the second part of the article for more on the impact of the Chinese hydro season). Based on these assumptions, we estimate the amount of electricity available for Bitcoin mining to be at least 9.6 GW.

Bitcoin hashrate and power consumption at its recent May 10 high and May 17 low.
: BitOoda, Blockchain.com, Kaiko, Coinmetrics.

According to our estimation, the bitcoin mining industryuses ~ 67% of the 9.6 GW of available electricity and this value is growing by ~ 10% per year. Most of today's devices are in the S17 class, but future growth will mainly come from the next generation of equipment, the S19 class. Some of the hashrate that returned to the network after May 17th was probably sourced from S9-class mining rigs, either located in extremely cheap jurisdictions, or with some delay (to avoid equipment downtime in the days leading up to halving) moving from the more expensive regions of northern China to the provinces Sichuan and Yunnan in order to take advantage of cheaper electricity during the annual flood season.

In addition, despite supply chain disruptions,by this time, limited shipments of the next generation equipment - Antminer S19 and Whatsminer M30 - had begun, as well as some shipments of the S17 class, thanks in part to which the hashrate was restored.

Hashrate, power consumption and installed base of Bitcoin mining equipment as of 07/01/2020.
: BitOoda, Blockchain.com, Kaiko, Coinmetrics.

About 50% of mining capacity is concentrated in China, another 14% in the United States.

Information about where the miningcapacity and how much miners pay for electricity, we drew from various open sources, as well as from confidential communication with miners and manufacturers and sellers of mining equipment. We were able to locate ~ 4.1 GW of power distributed among 153 miners, including 67 enterprises (with a total power consumption of ~ 3 GW) that provided price data on condition of anonymity.

Geographic distribution of the mining capacities studied in comparison with the total estimated energy consumption of 9.6 GW.
: BitOoda assessment, miners, ASIC manufacturers and sellers, public sources.

From information received by us in personalcontacts, one can come to the conclusion that most of the energy consumption is concentrated in the United States, Canada and Iceland, and only a relatively small part - in China and in the "rest of the world" category. We asked miners not only about their own capacities, but also about how many other miners in their market they know and what, in their opinion, the total amount of mining power consumption in their region. We understand how approximate the data obtained in this way is, but nevertheless we find this approach useful for assessing the overall geographic distribution of mining power.

Geographic distribution of the mining capacities studied in comparison with the total estimated energy consumption of 9.6 GW.
: BitOoda assessment, miners, ASIC manufacturers and sellers, public sources.

We estimate that 50% of the mining capacityBitcoin pays 3 cents per kWh or less, which is in line with the trend of a steady decline in this indicator over the past few years. Some evidence suggests that in 2018 this figure was closer to 6 ¢ / kWh. With a decrease in income per petahesh / sec. Due to the growth of the network hashrate, miners with high electricity costs were forced to either move to regions with lower electricity costs or shut down.

Electricity Cost Curve: Comparison of electricity costs with the share of the grid's energy consumption.
: BitOoda assessment, miners, ASIC manufacturers and sellers, public sources.

Based on our estimate of the cost curve, the averagethe monetary value of mining 1 BTC is around $ 5,000, with an upper confidence limit of around $ 6,000. This estimate includes cash operating expenses excluding depreciation or other costs of mining equipment.

The curve also shows that a small percentagebitcoins are mined at a cash cost above the current BTC spot price. We believe that some of this uneconomical mining is driven by power purchase commitments and potential incentive payments to shut down capacity during peak periods of power demand and bitcoin purchases in jurisdictions with limited or more costly trading opportunities.

Mining cost for 1 BTC based on network power consumption at various energy costs, data as of 07/01/2020.
: BitOoda assessment, miners, ASIC manufacturers and sellers, public sources.

Note that for breakeven mining onS9 hardware at the current network hash rate needs an electricity price of less than 2 ¢ / kWh, and it is likely that for such mining to remain viable, this upper price limit will need to decline even more over time as hash rates continue to rise. Our cost model assumes that one person is required to manage 5 MW of power consumption. Since S9 class devices are less energy efficient than newer mining rigs and require more devices per petahesh / sec. hashrate, they consume more energy compared to newer devices and require more labor and overhead to get the same hashrate. On S19 class equipment to generate 1 petahesh / sec. hashrate requires about 30 KW and just over 9 devices. On S9 class equipment to generate the same 1 petahesh / sec. the hash rate would require about 70 devices and more than 100 kW, and, accordingly, more labor and overhead costs.

Daily revenues and cash operating costs for various mining rigs at current hash rates and varying electricity costs.
Note: To estimate the share of electricity actually used for mining, we use a PUE of 1.12.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.
Labor cost is based on receivedestimates from miners regarding the cost of maintenance and operating personnel required to launch a large-scale (> 50 MW) project.

In summary, we estimate the available for miningBitcoin's energy capacity is ~ 9.6 GW, and the current utilization rate is 60% +. This electricity has an average price of ~ 3 ¢ / kWh, and the average cash cost of mining 1 BTC is about $ 5000. We estimate that China accounts for about 50% of total energy consumption, while the United States accounts for about 14%. Much of China's capacity migrates to Sichuan and Yunnan during the flood season to benefit from cheap electricity. This phenomenon will be discussed in more detail in the second part of the article.

Part 2: Some Surprising Findings on the Link Between Bitcoin's Network Hashrate Growth and China's Flood Season

We found that China accounts for 50% of thepower consumption of bitcoin mining and network hashrate. In this installment, we take a closer look at the Chinese Bitcoin community and the impact of the flood season on Bitcoin price and network hashrate.

What is the flood season? In the southwestern provinces of China Sichuan and Yunnanheavy rainfall falls from May to October. This results in a huge inflow of water to the dams, causing a surge in hydropower production. The production capacity of hydroelectric power plants during this period significantly exceeds the demand for electricity, so its surplus is sold to miners at a large discount. This frees up excess water from overcrowded dams, so selling cheap electricity is a mutually beneficial option for both hydroelectric power plants and miners. Surplus electricity at a cheap price attracts miners who transport equipment here from neighboring provinces. Whereas in northern China, during the dry months, miners pay 2.5-3 кВт / kWh for electricity, in Sichuan and Yunnan provinces during the rainy season, which lasts from May to October, the price drops below 1 ¢ / kWh.

We disagree with the conventional wisdom thatlow electricity prices stimulate network hashrate growth during the flood season. We believe the flood season is pushing the cost curve down 6 months a year, resulting in less forced Bitcoin sales needed to fund operating costs as miners accumulate capital to build capacity later.

As shown in the diagram below, there is a significantthe difference between the average price increase during the flood and dry seasons, while the hash rate growth during these periods remains approximately the same. We showed growth in each separate period, recognizing that the first two periods were likely to be exceptions (which further confirms our thesis), and the averages are based on a small sample of the subsequent 11 alternating 6-month periods.

BTC hashrate and price divided into dry and flood seasons.
Notes: data from 2014 are taken into account; the average does not include the period from November 2013 to October 2014; data as of 01.07.
: BitOoda, Blockchain.com, Kaiko, Coinmetrics.

This dynamic of capital accumulation, accompanied bythe purchase, supply and deployment of equipment is more broadly reflected in the correlation between the price increase (supported by capital accumulation) and the increase in hash rate 4-6 months after that when the purchased equipment is delivered.

China's flood season drives down the curvecosts, which can contribute to capital accumulation and future hash rate growth. Increased capital accumulation would reduce the industry's demand for external funding to support future hashrate growth.

Correlation between price and hash rate changes.
Note: data are given for the last 12 months, as of 01.07.
: BitOoda, Blockchain.com, Kaiko, Coinmetrics.

We consider the correlation of price changes infor a period of 15 to 360 days with hash rate changes over the same period last year. We noted that the highly correlated hash rate follows the price with a 4-6 month lag. This creates a dynamic of capital accumulation with the subsequent purchase, delivery and deployment of equipment as orders are processed in the supply chain.

Available and underutilized energycapacity, generation and capital accumulation within the industry (which is facilitated by the hydro season in China) and external financing, as well as a decrease in profit per petahesh / sec. - all this plays a role in the future growth of the hash rate. We will look at the future of hash in the third part of the article.

Part 3: Bitcoin Hashrate Growth Predictions: How Much, When, Why, and What Can Slow (or Accelerate) Growth

In this part of the article, we will talk about howthe hash rate of the network may grow, what factors contribute to its growth, as well as restrictions on the availability of capital and financing, which may restrain this growth.

According to our estimates, Bitcoin hashrate in the following12-14 months could exceed 260 exaches / sec, resulting in a slight increase in available power capacity from 9.6 to 10.6 GW and a renewal cycle that replaces older S9 equipment with S17 and S19 devices. The growth in power capacity is based on the available capacity at mining sites, planned infrastructure costs and the perception that in some locations with the highest mining costs, miners may be forced to close their businesses under pressure from dwindling revenues.

Bitcoin hashrate and power consumption.
Note: To estimate the share of electricity actually used for mining, we use a PUE of 1.12.
Data as of 01.07.2020.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

Completion of the equipment upgrade cycle beforeS19 class by mid-2022 will probably raise the network hash rate to ~ 360 exaches / sec. According to our estimates, the next radical upgrade of mining devices may not occur until mid-to-late 2022, although until that time we can expect a gradual increase in their energy efficiency. It is worth noting that if the bitcoin price stays at the same level or decreases, the gross income per petahesh / sec. will continue to decline to the level of marginal production costs, and further investment and hash rate growth in this case may slow down significantly, inevitably having an impact on the achievement of our predicted hash rate, which will be achieved later or not at all.

We tried to assess the progress of TSMC in comparison withSamsung and Intel (although Intel does not manufacture ASIC miners), and the available evidence suggests that there are large differences between the manufacturing processes of different semiconductor vendors. We anticipate that the next important step in ASIC technology will be associated with the transition to 5nm lithography. In this regard, TSMC, Bitmain's main supplier, is ahead of Samsung. However, although TSMC has bulky orders for both 7nm and 5nm functional nodes, their geometry is very similar to Intel's 10nm nodes, and we believe that Samsung, by making more winning geometry, literally breathes into the back of TSMC. ASICs are primarily logic chips, so a comparison to Intel makes sense. As the semiconductor industry develops, we notice an increasing divergence in element geometry. Thus, there are important differences with respect to chip density, cell sizes, and ultimately power consumption and thermal properties of chips from different manufacturers, even in the same nominal functional unit.

Comparison of the geometry of Intel and TSMC chips.
: https://www.eetimes.com/intels-10nm-node-past-present-and-future/

Samsung recently announced (eng.) that the planned commercial production at 3nm functional nodes is likely to be postponed until 2022, and the core of production in 2021 is likely to be 5nm nodes. We believe that a shortage of 3nm hardware is likely to lead to 5nm nodes forming the backbone of ASIC development and production by 2022. For these reasons, we expect S19 equipment to make up the majority of shipments over the next 24 months, although incremental design improvements could lead to efficiency gains, which may be reflected in new product lines.

The hash rate growth (bottom) slows down after using the available power capacity and completing the update cycle.
Note: To estimate the share of electricity actually used for mining, we use a PUE of 1.12.
Data as of 01.07.2020.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

As the graphs above show, evenRelatively modest assumptions about the growth of the energy capacity of the network and the widespread deployment of mining equipment of the S19 class result in a hash rate of 360 exaches / sec. Improving energy efficiency (fewer watts per thesh / sec.) May be a positive factor for these predictions, but the critical issue is the decrease in BTC received, per petahesh / sec. or per MWh, the total daily flow of bitcoins remains roughly constant, fluctuating only with additional blocks and transaction fees. So, if the network hashrate increases, then the share of miners in the total hashrate and, therefore, in the BTC stream decreases. And if the price of BTC does not grow in line with the growth of the hash rate, mining profitability will fall, and a new equilibrium can be achieved at a significantly lower hash rate than what we are predicting here.

The chart below shows the bitcoins earned per MWh for each device class: S19 miners can earn almost 3 times more BTC per MWh than S9.

The amount of BTC earned per MWh depending on the network hashrate.
Data as of 04/07/2020; Energy figures are based on PUE 1.12.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

The diagram below shows how (and,presumably will change) the amount of BTC received per petahesh / sec. depending on both the network hashrate and the time, and taking into account the block reward before and after the halving. And here, too, you can observe a decrease in profitability in BTC. When looking at this issue regardless of device, based on petahesh / sec alone, it is clear that price is a key element in sustainable hash rate growth over time.

Daily volume of bitcoins earned on petahesh / sec. over time and depending on the network hashrate.
Historical data starting from 01.01.2018, as of 01.07.2020.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

The dollar value of the mined BTC will fall over time, leading to a decrease in the profitability of mining, If only the price of bitcoin will not rise enough for thiscompensate. As shown in the diagram below, the income generated per petahesh / sec is a function of both the network hashrate and the bitcoin price. With the current target network hash of ~ 124 exaches / sec. and a price of $ 9220 / BTC, daily income per petahesh / sec. is ~ $ 70. If the network hashrate grew to 260 exahesh / sec, which, as we expect, should happen in the summer of 2021, then to get the same daily income of $ 70 per petahesh / sec. bitcoin would have to be worth about $ 19,500. At a price of $ 10,000 / BTC, daily income per petahesh / sec. will be only $ 36. The middle chart below shows that mining costs with efficient S19 hardware and an electricity price of 4 ¢ / kWh are ~ $ 37 per petahesh / sec. per day, but mining on S9 class equipment at the same electricity price will cost $ 133 / day. To breakeven mining on S9 class equipment at a BTC price of $ 10,000, electricity must cost 0.5 ¢ / kWh.

Daily income and cash operating costs for each class of mining equipment at different cost of electricity and with several options future hashrate and BTC prices.
Note: To estimate the share of electricity actually used for mining, we use a PUE of 1.12.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

Significant capital expenditures required forreaching the potential hash rate is a limiting factor - in particular, if the rise in the BTC price does not keep pace with the hash rate, this, at least, will hinder the generation of capital within the industry, further increasing the dependence on external sources of capital. In addition, it could downgrade our hashrate forecasts, displacing more expensive equipment and limiting external capital flows when projects face uncertainty and lower investment return expectations.

What if the price moves horizontallycorridor? At what point will the hash rate stop growing? With an electricity price of 1 ¢ / kWh, S9 class equipment could continue to operate with a network hash rate of up to 180 exaches / sec. With a price of 3 ¢ / kWh, S19 devices can operate at hash rates up to 295 exaches / sec. With a higher hashrate, breaking even S19 mining will require either a higher BTC price or a lower electricity price. However, if the hash rate is well below 295 exaches / sec., Then the devices will not be able to recoup the capital expenditures for their use. Obviously, the rise in the BTC price is included in the capital investment budget of each miner.

Daily revenue and cash operating costs per petahesh / sec. depending on the network hashrate.
Note: To estimate the share of electricity actually used for mining, we use a PUE of 1.12.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

The total capital investment required forreaching 260 exaches / sec. over the next 12 months is $ 4.5 billion, plus ~ 2 billion to reach 360 exaches / sec. by mid-2022.

Bitcoin hashrate and power consumption.
Note: To estimate the share of electricity actually used for mining, we use a PUE of 1.12.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

If the price of bitcoin within two years isgrow steadily to ~ 19 thousand dollars at a rate of 40% or more per year, then S19 class equipment will remain viable even with electricity at 5 ¢ / kWh, however, there will be a gap of ~ 4.1 billion dollars between the total demand industry in capital investments and cash flow generated within the network.

The capital cost of the Bitcoin network and the cash flow generated within it.
Data as of 01.07.2020; logarithmic scale on the Y-axis.
: BitOoda assessment, Blockchain.com, Kaiko, Coinmetrics.

We have had to face concerns aboutthe fact that our model of hashrate growth implies a significant supply of new equipment: we were asked how feasible this forecast is. For the installed and connected mining equipment base to grow in line with our forecast, about 60,000 devices should be shipped weekly. By comparison, Bitmain shipped 95,000 pieces of hardware per week in the first half of 2018, according to company reports. Despite some uncertainty regarding the number of chips and die sizes in the S19 class miners, we are confident that semiconductors and production capacity will not be a limiting factor.

In conclusion, we believe that the network hashrateBitcoin can reach 260 exaches / sec. for 12 months and 360 exaches / sec. within 24 months. However, this depends to some extent on the price of bitcoin, which, according to our model, should rise by about 25-35% per annum. We do not model or predict the future price of Bitcoin, but only reflect the impact of potential price scenarios on hashrate growth, energy consumption and capital costs of the mining industry, and mining profitability. A deviation from the designated range can delay or, conversely, accelerate the growth of the hash rate. The price of bitcoin and the availability of outside capital to bridge the funding gap could potentially limit the industry's ability to increase mining capacity to 360 exaches / sec, but the ability to manufacture or assemble the required semiconductor chips is not.

Investors when evaluating mining projectsone should take these predictions into account and not forget about the impact of the bitcoin price. We recommend that investors stick to an active hedging strategy to mitigate operational risks - as we like to say, miners know what their costs will be in 6, 12 and 24 months, but they don't know how much bitcoins they will receive or how much those bitcoins will cost. Hedging strategies can help reduce operational risks and stabilize cash flows.

</p>