Understand the world's Bitcoin mining costs and Bitcoin computing power in one article丨Mining Report

Mining is a fundamental component of the Bitcoin network and BTC as an asset. Despite its importance, mining has become the least transparent and least understood part of the broader Bitcoin ecosystem. This report from BitOoda aims to increase transparency related to the composition of Bitcoin miners, which ultimately helps understand the state and health of the system. At the Fidelity Center for Applied Technology (FCAT), we look forward to continuing our research into the Bitcoin mining space and helping to advance the ecosystem. We appreciate the work done by the BitOoda team, which we hope will increase everyone’s understanding of this complex and fascinating part of the Bitcoin ecosystem.

——Juri Bulovic, Fidelity Applied Technology Center

Key points:

• 50% of the world’s 9.6 GW of mining power may be located in China; the United States accounts for about 14%; and the computing power utilization rate is about 67%.

• The median Bitcoin computing power cost is 3 cents/kwh, and the mining cost of 1BTC is about $5,000.

• In one upgrade cycle, hashrate could reach 260 EH/s in 12 months and 360 EH/s in 24 months, but would require $6.3 billion in capital expenditures; leaving a funding gap of $4.1 billion.

• For every 10EH/s increase in hashrate, the Bitcoin price needs to rise by $1,000 to maintain the median MWh revenue.

• Cheap electricity, bitcoin prices and semiconductor allocations are all factors

Bitcoin mining is a secretive industry with little public information. We found that even sophisticated cryptocurrency investors have gaps in their understanding of mining and potential investment opportunities in the space. Despite excellent research by Coinmetrics, Coinshares, and the Cambridge Center for Alternative Finance, there are still unanswered questions. The research for this article has been commissioned by the Fidelity Center for Applied Technology (FCAT) to add to existing research and attempt to address new questions.

Part 1: Bitcoin hashrate analysis: how much is it, where does it come from, and how much does it cost

In the first part, we attempt to measure, locate, and price miners’ hashrate, and estimate miners’ profitability. Through more than 60 conversations with miners, mining machine manufacturers, and resellers, as well as more than 45 public data sources, we strive to provide as complete a picture as possible of how much Bitcoin mining hashrate there is, where it is located, and how much it costs miners to mine.

We then further explore the question of how future mining capacity will vary with available electricity, the efficiency of mining equipment and the price of Bitcoin, capital/funding availability, and possible limitations on semiconductor technology and production capacity.

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

The 9.6GW estimate uses the following logic: On May 10, just before Bitcoin’s third block reward halving, the network hashrate peaked at 136,098 PH/s and bottomed out at 81,659 PH/s on May 17. We acknowledge that some of these extremes may be due to luck — such as a lucky streak or finding blocks quickly — which could artificially inflate the estimated hashrate, and that slower block rates may be partly due to bad luck. However, we exclude luck from our model and make simplifying assumptions to arrive at an approximation of how much power the Bitcoin network consumes. We assume that all of the hashrate at the May 17 trough came from the more profitable new generation of mining equipment, the “S17 class” of miners, which includes Bitmain’s Antminer S17, T17, Whatsminer M20, and equipment from Canaan Creative, Innosilicon, Ebang International, and other manufacturers. We also assume that all equipment that was shut down between May 10th and May 17th was of an older generation, such as the less profitable S9 category of miners (e.g. Antminer S9, Whatsminer M3).

Note that we use “S17 Class,” “S9 Class,” and “S19 Class” as generic terms to include Bitmain equipment as well as competing equipment with similar specifications. Because Bitmain has a long history of market dominance in the “S9 Class” (and to a lesser extent in the “S17 Class” equipment), we use only Bitmain models to define the categories. In all relevant calculations, we also assume a PUE (power usage effectiveness) of 1.12, meaning that for every 1MW used to directly mine Bitcoin, 120kw will be used to run everything else, including cooling systems, lighting, servers, switches, etc.

Figure: BitOoda classifies mining machines into key mining machine categories

Sources: BitOoda, Bitmain, Canaan, MicroBT, Halong, GMO, AsicMinerValue.com

The chart below shows that if all the hashrate running on May 17th was from newer S17 class equipment, they would have consumed 3.9GW of energy. Additionally, if the 54EH/s of hashrate that was shut down between May 10th and May 17th was from older S9 class rigs, that accounts for another 5.7GW of power. We make these simplifying assumptions to help give a broad understanding of the industry, knowing that the likely reality is that most, but not all, of the hashrate that was shut down were older S9 class rigs, and that a small portion of the remaining hashrate was likely S9 class rigs in very low-cost power markets. The decline in profitability of mining equipment after the Bitcoin halving was a key driver of the hashrate decline, coupled with some timing related to moving miners from northern China to southern China to take advantage of lower power costs (see Part II for more details on the impact of China’s hydropower season). Based on these assumptions, we estimate that the Bitcoin mining community has access to at least 9.6GW of power.

Figure: Bitcoin hashrate and energy consumption at recent peaks and troughs

Source: BitOoda, Blockchain.com, Kaiko, Coinmetrics

We estimate that BTC mining utilizes 67% of the 9.6GW and is growing at about 10% per year, powering 2.8 million dedicated Bitcoin mining rigs. Most of the current equipment is S17 class, but future growth will come primarily from next generation S19 class mining equipment. Some of the hashrate that has risen from the May 17 trough may come from S9 class rigs, which are either operating in jurisdictions with very low electricity costs, or from higher cost regions in northern China that were previously delayed in moving (to avoid downtime before the May 11 reward halving) and are now taking advantage of low cost electricity in Sichuan and Yunnan during China’s annual flood season.

Furthermore, despite supply chain delays, limited shipments of the next-generation Antminer S19 and Whatsminer M30 have begun, along with some shipments of the S17 class of miners, driving the recovery in hashrate.

Figure: Bitcoin hashrate, energy consumption, and number of mining machines installed — data as of July 1, 2020

Source: BitOoda, Blockchain.com, Kaiko, Coinmetrics

China is estimated to account for about 50% of mining capacity, while the United States accounts for another 14%.

We used a variety of public sources, as well as confidential conversations with miners, mining equipment manufacturers, and resellers to assess where Bitcoin mining power is located and how much miners pay for electricity. We were able to find approximately 4.1GW of power in 153 mining sites, including 67 mining sites, or approximately 3GW of power capacity, that provided electricity price data on the condition of anonymity.

Figure: Geographical distribution of mining capacity surveyed vs. estimated 9.6GW of electricity

Source: BitOoda estimates, miners, ASIC manufacturers/distributors, public sources

Our conversations lead us to believe that we account for the majority of capacity in the United States, Canada, and Iceland, but only a small percentage in China and the “Rest of World” category. In our discussions with miners, we ask not only about their own hashrate, but also how many other miners they know in the market and how much total hashrate they think the region has. We know these are approximate, but we still find this to be a useful way to estimate the overall geographic distribution of mining power.

Figure: Geographical distribution of measured computing power and estimated total capacity of 9.6GW

Source: BitOoda estimates, miners, ASIC manufacturers/distributors, public sources

In our assessment, 50% of Bitcoin mining electricity pays 3 cents/kWh or less, a figure that has continued to decline over the past few years. Evidence suggests that this figure was closer to 6 cents/kWh in 2018. As revenue per PH/s of hashrate declines as network hashrate increases, miners with high electricity costs either move to lower-cost regions or shut down.

Figure: Power cost curve: mapping the relationship between power cost and network power share

Source: BitOoda estimates, miners, ASIC manufacturers/distributors, public sources

Our cost curve estimate translates to a median cash cost of about $5,000 to mine 1 Bitcoin , with an upper confidence interval of about $6,000. This estimate is a cash operating expense and does not include depreciation or other expenses for mining hardware.

The curve also shows that a small portion of Bitcoin is mined at a cash cost above the current Bitcoin spot price. We believe that some of this uneconomic mining is governed by power purchase commitments and potential incentives to shut down capacity during peak demand periods and acquire Bitcoin in jurisdictions where trading options are limited or expensive.

Figure: Cost to mine 1 BTC based on network hashrate at different electricity costs — data as of July 1, 2020

Source: BitOoda estimates, miners, ASIC manufacturers/distributors, public sources

We note that S9-class miners need less than 2 cents/kWh to break even at current network hashrate, and may need even lower electricity prices to remain viable as hashrate continues to grow. Our cost model assumes that one person runs about 5MW of electricity. Since S9-class devices are less energy efficient than newer miners and require more devices per PH/s of hashrate, they consume more energy than newer devices and require more labor and overhead costs for the same hashrate. S19-class miners require 30kW of power and more than 9 devices to generate 1PH/s of hashrate. If mining with S9-class miners, it would require about 70 devices and more than 100kW, and correspondingly, the labor costs and electricity overhead for maintenance and operation would increase accordingly to generate the same 1PH/s of hashrate.

Figure: Daily revenue and cash operating costs of different mining machines at different electricity prices at current hash rates

Note: We assume a PUE of 1.12 to estimate the share of electricity actively used to mine Bitcoin

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

According to our miner conversations, labor costs are based on the maintenance and operations staff required to run a large-scale mining farm (> 50MW).

Summary: We estimate that there is about 9.6GW of available power capacity for mining Bitcoin, with current utilization in the mid-60% range. The median price of this power is about 3 cents/kWh, and the median cash to mine 1 BTC is $5,000. We estimate that China accounts for about 50% of global capacity, with the United States close behind at about 14%. The relocation of a large portion of China's power capacity during flood season to take advantage of lower power prices is detailed in Part II.

Part 2: Some surprising conclusions about the relationship between computing power growth and China’s flood season

We found that China contributes 50% of Bitcoin mining energy consumption and network hashrate. Here we take a closer look at the Chinese Bitcoin community and the impact of China’s hydropower peak season on Bitcoin price and hashrate network.

So what is hydropower season? From May to October, China’s southwestern provinces of Sichuan and Yunnan face heavy rainfall. This causes dams in these provinces to fill up, resulting in a surge in hydroelectric power during this period. This power is sold cheaply to Bitcoin miners as production capacity exceeds demand. Overflowing dams release excess water, so selling cheap power is a win-win for both utilities and miners. This cheaper power price attracts miners from nearby provinces to migrate over to take advantage of the low power prices. During the dry months, miners in North China pay around 2.5–3 cents/kWh for electricity, while miners in Sichuan and Yunnan pay less than 1 cent/kWh during the rainy season from May to October.

We argue against the conventional wisdom that low electricity prices drive hashrate growth during the flood season. We argue that the flood season pushes the cost curve downward for six months of the year, causing miners to accumulate capital to support hashrate growth and reduce the selling of Bitcoin to cover operating expenses.

As shown in the chart below, there is a meaningful difference between the average price increases during the wet and dry periods, while the growth rate of hashrate is roughly the same during both periods. We recognize that the first two periods are likely outliers (further supporting our thesis), and each period shows growth, and the average is based on a small sample size of 11 alternating 6-month periods that followed.

Figure: Hashrate and BTC price during flood and dry seasons

Note: Since 2014; average excludes November 2013 to October 2014; data as of July 1, 2020

Source: BitOoda, Blockchain.com, Kaiko, Coinmetrics

The correlation between BTC price increases (which underpin capital accumulation) and hashrate growth more broadly reflects the dynamics of capital accumulation, subsequent equipment purchase, delivery and deployment, which in turn underpins hashrate growth 4-6 months later as the supply chain gradually delivers the purchased mining equipment.

China’s flood season could lead to a lower cost curve, which could help capital accumulation and help promote future computing power growth. Increased capital accumulation would reduce the industry’s need for external funding to support future computing power growth.

Figure: Correlation between price changes and computing power changes

Note: Past 12 months, data as of July 1, 2020

Source: BitOoda, Blockchain.com, Kaiko, Coinmetrics

We look at the correlation between price changes over the past 15 to 360 days and hashrate changes over the same period last year. We notice that hashrate follows price, with a lag of 4-6 months, and with a high correlation. This sets up a dynamic of capital accumulation, followed by equipment purchase, delivery, and deployment, as the supply chain delivers the purchased equipment.

Available and underutilized generation capacity, capital accumulation for generation within the industry (affected by China’s hydropower season), external funding, and reduced revenue per PH/s all have an impact on the future growth of hashrate. We explore the future of hashrate in Section 3.

Section 3: Bitcoin hash rate growth forecast: how much, when, why and what factors will slow (or accelerate) growth

We will take a closer look at how much hashrate can grow, what factors support this growth, and the funding and capital constraints that may mitigate this expected growth.

Based on our assessment, the Bitcoin network can exceed 260EH/s hashrate in the next 12-14 months with a modest increase in available power capacity from 9.6GW to 10.6GW and a mining machine upgrade cycle that replaces older S9-class machines with newer S17 and next-generation S19-class machines. The growth in power capacity is based on available power at mining sites, planned infrastructure spending, and the view that some higher-cost mining sites may need to shut down operations due to revenue pressures.

Figure: Bitcoin computing power and energy consumption

Note: We assume a PUE of 1.12 to estimate the share of electricity actively used to mine Bitcoin

Data as of July 1, 2020

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

Completing the upgrade cycle to S19-class miners by mid-2022 could bring the network hashrate to ~360EH/s. We estimate that the next fundamental equipment upgrade is unlikely until mid-to-late 2022, although power efficiency improvements are expected in the interim. We note that if the Bitcoin price remains constant or declines, then USD revenue per PH/s will continue to decline to the marginal cost point, and further investment and hashrate growth could slow significantly - so our hashrate forecasts could face delays, or never be achieved.

We observed TSMC’s progress compared to Samsung and Intel – although Intel does not produce ASICs, available data indicates a large difference between process technologies of different semiconductor suppliers. We note that the next major advancement in ASIC technology will be the development of 5nm technology. At this node, TSMC (Bitmain’s main supplier) is ahead of Samsung. However, although TSMC has seen volume orders at both the 7nm and 5nm nodes, their process geometries look similar to Intel’s 10nm node. We believe Samsung also has tighter process geometry features; therefore, Samsung is close behind TSMC. ASICs are primarily logic chips, so it makes sense to compare with Intel. As the semiconductor industry evolves, we notice that there are increasing differences between feature geometries, so even at the same nominal process node, there are important differences between chip density, feature size, and ultimately power consumption and thermal performance between different chip manufacturers.

Figure: Comparison of Intel and TSMC process geometry

Source: https://www.eetimes.com/intels-10nm-node-past-present-and-future/

Samsung recently announced that plans to commercially produce on the 3nm process node may be delayed until 2022, with 5nm likely to become the bulk of production in 2021 (see this news article). We believe that the lack of 3nm capacity and likely low initial yields will result in 5nm becoming the primary means of ASIC development and production until 2022. For these reasons, we believe that S19-class miners will account for the majority of shipments over the next 24 months, although incremental design improvements could lead to efficiency gains that could be reflected in new model lines.

Figure: Once power capacity is utilized and upgrade cycles are complete, growth in hashrate (bottom) slows down

Note: We assume a PUE of 1.12 to estimate the share of electricity actively used to mine Bitcoin, data as of July 1, 2020

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

As shown in the chart above, even relatively modest assumptions about the growth of network power capacity and widespread deployment of S19-class mining machines can get us to a network hashrate of 360EH/s. Higher energy efficiency (fewer watts per TH/s) could give these projections a boost, but a key issue is that the number of Bitcoin earned per PH/s or per MWh is decreasing - the total Bitcoin flow per day remains roughly constant, fluctuating only with new blocks and transaction fees. Therefore, if the network hashrate increases, the miners' share of the total hashrate decreases, resulting in a decrease in Bitcoin flow. If the price of Bitcoin does not keep up with the growth in hashrate, then profitability will decrease and the new equilibrium of hashrate could be significantly lower than we expect.

The chart below shows the Bitcoin earned per MWh for each device category: The S19 can earn almost three times more BTC per MWh compared to S9 class devices.

Figure: Bitcoin earned per MWh of computing power generated

Data as of July 4, 2020; energy consumption figures include a PUE of 1.12

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

The chart below shows how BTC per PH/s has changed over time with network hashrate (and is expected to change in the future), accounting for halvings and block rewards. Here, one can also see the decline in revenue in BTC terms. Viewed on a per PH/s basis, regardless of device, it is clear that price is a key factor in the continued growth of hashrate over time.

Figure: Daily amount of Bitcoin earned per PH/s and network hashrate over time.

Data as of July 1, 2020; historical data since January 1, 2018

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

The dollar value of mined BTC will decline over time, resulting in declining profitability, unless the price of Bitcoin rises enough to offset this loss. As shown in the chart below, the revenue earned per PH/s per day depends on the network hashrate and the Bitcoin price. At the current target network hashrate of ~124EH/s and the current BTC price of $9220, the daily revenue per PH/s is about $70. If the network hashrate increases to 260EH/s, as we expect in the summer of 2021, the Bitcoin price would need to reach $19,500 for the daily revenue per PH/s to remain the same at $70. At a BTC price of $10,000, the revenue per PH/s is only $36. The middle chart below shows that running 1PH/s per day on an efficient S19 class miner at an electricity cost of 4 cents/kWh costs about $37, but running an S9 class rig at 4 cents/kWh would cost $133. An S9 rig would need to operate under 0.5 cents/kWh to break even at a BTC price of $10,000.

Figure: Daily revenue and cash operating costs for each mining machine category at different future prices and at different future prices.

Note: We assume a PUE of 1.12 to estimate the share of electricity used to mine Bitcoin

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

The significant capital expenditures required to realize potential hashrate is a limiting factor - especially if the BTC price does not rise to keep pace with hashrate, it will at least suppress the industry's internal cash flow, further increasing its reliance on external capital. In addition, as proposed projects face uncertainty and reduced expectations of return on investment, this could cause higher-cost miners to exit operations and limit external capital flows, which would adversely affect our hashrate forecasts.

What if the price remains constant? At what point does hashrate stop growing? If the price of electricity is 1 cent/kWh, an S9-class miner can keep up to 180EH/s of network hashrate running. At 3 cents/kWh, an S19-class device can maintain up to 295EH/s of network hashrate. Beyond that point, an S19-class miner would need either a higher BTC price or lower electricity prices to keep running. However, these devices would not be able to recoup their capital costs at hashrates much below 295EH/s. Clearly, price appreciation is baked into every miner’s capital budget.

Figure: Daily revenue and cash operating costs per PH/s vs. network hashrate

Note: We assumed a PUE of 1.12 to estimate the share of electricity actively used to mine Bitcoin

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

The total capital expenditure required to reach 260 EH/s in the next 12 months is $4.5 billion, and an additional $2 billion is required to reach 360 EH/s by mid-2022.

Figure: Bitcoin computing power and energy consumption

Note: We assumed a PUE of 1.12 to estimate the share of electricity actively used to mine Bitcoin

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

If the Bitcoin price appreciates steadily at a rate of more than 40% per year to around $19,000 within two years, then even at an electricity cost of 5 cents/kWh, S19-class mining machines will still be viable, but the total funding gap for the industry's capital expenditure needs and cash flow generated within the network will reach approximately $4.1 billion.

Figure: Bitcoin network capital expenditure and internal cash flow

Data as of July 1, 2020; logarithmic on Y axis

Source: BitOoda estimates, Blockchain.com, Kaiko, Coinmetrics

We have raised concerns that our hash rate growth model requires a large number of new miners to be delivered, and have received questions about whether our forecast is feasible. Approximately 60,000 units would need to be shipped per week to increase the installed base of mining equipment to be consistent with our hash rate forecast. In comparison, Bitmain was able to deliver over 95,000 S9 miners per week in the first half of 2018, according to company filings. While we are uncertain about the number of chips/die sizes inside the S19 miners, we are confident that semiconductor/assembly capacity will not be a limiting factor.

In summary, we believe the Bitcoin network hashrate could reach 260 EH/s within 12 months and 360 EH/s within 24 months. However, this is somewhat dependent on Bitcoin’s price appreciation or expected appreciation at a rate of 25–35% per year, according to our model. We do not model or forecast future prices for Bitcoin, but only reflect the impact of potential price scenarios on hashrate growth, electricity consumption, and mining industry capital investment and profitability. Differences in this range could delay or accelerate hashrate growth. The price of Bitcoin and the availability of external funding to bridge the funding gap are potential constraints on the industry’s ability to increase Bitcoin mining capacity to 360 EH/s, but the ability to produce or assemble the required semiconductor chips is not.

Investors need to consider these forecasts when evaluating mining projects and need to understand the price of Bitcoin. At BitOoda, we are strong supporters of hedging and recommend that investors adopt an active hedging strategy to reduce operational risk - as we like to say, miners know what they will spend in 6, 12 and 24 months, but they don’t know how much Bitcoin they will receive, or how much Bitcoin will be worth. Hedging strategies can help reduce operational risk and stabilize cash flow. Please contact us at [email protected] to receive a full report on the methodology, resources and more details provided in this article, or to discuss risk management strategies and trading opportunities we can work with you.

About BitOoda:

BitOoda is a fintech and financial services company focused on digital assets. We have been a pioneer in building the BTC and ETH derivatives market and have helped companies develop, manage and execute risk management strategies through structured derivatives and proprietary investment products.

Link to this article: https://www.8btc.com/article/624178
Please indicate the source of the article when reprinting