The Evolution of the Bitcoin Mining Industry: January 2015 — Now
Mining Technology
Since the analysis undertaken in 2015, Bitcoin mining technology has improved dramatically. The benchmark used back then was Bitmain’s Antminer S5. We will look at the S5 compared to its current successor, the Antminer S9i[i].
Further to the above, one of Bitmain’s closest competitors, Canaan Creative, comes in with a lower $/GH rate ($0.044) when excluding PSU costs from both rigs, but a 15% higher W/GH value (0.109)[ii]. As the market will tend to gravitate towards the lowest total price available, it’s expected that Bitmain controls and ships significantly more hardware than Canaan[xxiv].
Hash Rate Growth
The dramatic drop in $/GH and W/GH shown in Table 1 has spurred extraordinary hash rate growth. That said, this is not a new phenomenon.
One of the main drivers of investment in mining equipment is expected hash rate growth from one difficulty cycle to the next. Table 2 shows how consistent fortnightly hash rate growth has been over the past 6 and a half years. Network difficulty grows directly in line with hash rate growth.
As a result of the constant hash rate increases, the difficulty cycle is rarely 14 days, and based on rough year to date data (7.2% increase per cycle), the difficulty cycle is closer to 14 days x (1–7.2 %) = 13 days, or 312 hours.
Should Bitcoin ever scale and reach its potential, it is almost certain that mining equipment will exponentially increase in processing efficiency in line with Moore’s Law for at least another 5 years[iii] and exponentially increase in power efficiency in line with Koomey’s Law for at least another 25 years[iv].
Understanding the Cost of Bitcoin — Economic Inputs & Drivers
Now that the evolution of network hash rate growth has been demonstrated, calculating the costs of Bitcoin can be modelled quite simply through the relationship of the 7 variables defined below.
I & II— HASH RATE & CAPEX
CAPEX is the capital expenditure required to maintain a proportional share of mining rewards upon an increase in difficulty. This is typically the purchase of additional GH/s at a particular $/GH rate. This is demonstrated in the below example, assuming the average of 7.2% difficulty increase discussed above:
Therefore, for the example mining pool to maintain their 30% slice of the pie, they need to bring on 21.6PH/s of hash power.
There are other elements of CAPEX whose life-cycles are much longer than mining equipment. These elements of CAPEX can also be deemed as “sunk costs” in many cases, and don’t affect future decisions. The CAPEX categories are as follows:
- Bitcoin Mining Equipment (typically last for only a few months before they’re unprofitable)
- Power Supply Units (PSU) for mining equipment typically last as long as the mining equipment due to planned obsolescence, with hardware manufacturers regularly changing the required PSU wattage with each new generation of miner.
- Server Racking / Data Centre Construction & Fitout Costs (typically last for decades). Server Racks / Data Centres could also come under Operational Expenses (OPEX) if the Data Centre is being rented / leased. Regardless, these costs are negligible compared to the costs of electricity.
III — OPEX
OPEX is the expenditure required to remain operational. At scale, this is effectively just the cost of power to the mining equipment and air conditioning within a data centre. It is estimated that cooling can consume 30[v] to 40%[vi] of overall energy consumption, with 21% a benchmark for the most efficient cooling systems[vii]. Technologies such as immersive cooling will reduce energy consumption as a trade-off for a large initial capital outlay. One should take in account the “Iceland Factor”, where Bitcoin mining uses as much power as all of Iceland’s homes[viii] due to it being cold enough for data centres to meaningfully reduce cooling costs and having clean and cheap hydro-electricity. At 840 GWh/yr, tiny Iceland would account for about 1% of the world’s mining power. While Iceland is only a very small share of the market, miners have access to several other cold places with cheap electricity[ix]. For the purposes of this model, we will assume cooling contributes to 20% of the total power consumption, in line with the laws of perfect competition and the technological gravitation towards maximum efficiency.
IV — Difficulty Cycle Length
As the network hash rate increases 7.2% on average, blocks will be mined, on average, 7.2% quicker. Therefore, a more accurate figure to use would be (1–7.2%) x 14 days, or, a period of 13 days, or 312 hours.
V — Coins Mined
This is a fixed number — there are 2016 blocks of 12.5 bitcoins mined every difficulty cycle — 25,200 bitcoins. In addition to the mining rewards, mining fees are not insignificant either[x]. The SegWit Wars of the first half of 2017 had fees averaging over 200BTC per day, and the fee madness during the hype cycle of December 2017/ January 2018 had a revenue average of over 550BTC per day over those two months. With the SegWit wars over, and the hype now well settled, a relatively consistent 40 BTC per day has been earned in the 6 months leading to July 31, 2018 (st dev = 35, n=180). Daily average fee revenue trends over time are shown in the table below. For this model, we will use a figure of 650 BTC collected in fees each cycle (i.e. about 50/day for the average 13-day cycle time).
V I— Power Cost & Emissions
To evaluate power costs and emissions, we don’t have much of a choice but to use world-wide weighted average figures, due to the dispersion of miners all over the world. That said, thanks to the rules of perfect competition, particularly perfect factor mobility, miners will move to places with the cheapest electricity costs. The statistics are as follows[xi],[xii],[xiii],[xiv]. The emissions figures consider CO2 equivalents, such as methane, and nitrous oxide.
*Note: When using Carbon Capture Systems (CCS), CO2 emissions from Coal are reduced substantially.
Although the average rate for US industrial companies is about $0.07/kWh[xv], a safer assumption for Bitcoin miners would be closer to 3 or 4 cents, for the reasons mentioned above. There are several documented cases of the largest bitcoin mining operations paying $0.04/kWh[xvi], with reports that Bitmain was receiving a $0.02/kWh rate in their Yunnan facility[xvii], with one particular CEO claiming a cost of electricity of only 1.7 cents/kWh for their mining operation in Moses Lake, Washington, USA[xviii].
VII — Mining Mix — “The Network Average Miner”
There are two types of miners; chip-fabricator miners, and retail miners. Retail miners can be split further into another two categories, large retail miners and small/individual miners. Small individual miners can also forego buying hardware themselves, and instead purchase mining contracts. Due to intellectual property and some economies of scale, chip-fabricators (chipfabs) can mine for significantly cheaper than retail miners. Typical gross profit margins in the semiconductor industry has averaged over 45% for a four-year period[xix], with the most profitable ones close to 60%. The computer hardware industry averages around 35%[xx]. Gross profit margins are used since operating expenses and depreciation are dealt with separately within the model. It is assumed that miners pay no tax (i.e. they retain all coins that are mined and/or asset depreciation costs are high enough to offset a large amount of tax on revenue from sold mining hardware). Due to the lack of competition in the ASIC hardware space, margins would likely be 50 to 60%. Obviously, there is a limit to the margin that can be made on mining hardware, as the customer base is quite savvy and can easily calculate profitability of their purchased miners at a particular price-point. For the purposes of this study, it will be assumed that Bitcoin ASIC manufacturers make 60% gross margin on all hardware sold.
Determining the number of non-chipfab large miners and individual miners is another area of speculation due to lack of robust market data. One half-insight can be gained from looking at the world’s largest cloud-mining operation, Genesis Mining, who claim to have 2 million users[xxi]. Despite its higher price per GH/s (16c/GH (including electricity costs and incidentals)[xxii] vs Bitmain’s 5.4c/GH), it is still practical for many miners to opt for a cloud-based solution due to its “plug-and-play” nature, and more importantly, that it is an “instant-on” solution, so that you don’t lose your most profitable days waiting for your miner to be shipped to you. That said, Genesis provides no data on their aggregate hash power, nor do they reveal details on the location of their server farms, or even which pools they mine on[xxiii], however they claim that “we will at least use two different pools (in some cases we use up to four) for each coin. This is to preserve the decentralized nature of the crypto networks! If we become aware that a pool is getting close to 50% share, we will switch away from it and use a backup instead.”
Next comes the question of chipfabs mining on their own equipment, and how much equipment has made it out into the market for large-scale and small-scale miners. According to an analysis by Sanford C. Bernstein & Co, it was estimated that Bitmain captured 75% of market share in hardware sales, Canaan Creative captured 15% of the market, and other manufacturers made up the remaining 10%[xxiv]. Bitmain’s CEO has stated that the company earned USD$2.5B in revenue for 2017[xxv], with the majority of that revenue earned through mining sales, as opposed to mining and selling Bitcoin directly. From this, we can size the market for mining hardware to be a maximum of USD$3.33bn. Based on the 2017 average price of an S9 miner of around USD$3000[xxvi], one could draw the conclusion that about 1 million S9-equivalent mining units were shipped. This estimate was confirmed on August 22, 2018 through Bitmain’s translated IPO deck, who have declared sales of 1 million S9 units in 2017[xxviii]. How’s that for good estimating?!
At this hash rate and price per S9i, this model estimates that roughly $115 million is invested in more mining power every difficulty cycle, or around $3.25 billion per year (slightly lower than the 2017 figures). Drawing on the 80/20 rule again we can put chipfabs somewhere in the ballpark of 20% of direct hash power. That said, with Bitmain administering at least two mining pools (AntPool & BTC.com) 24 providing 40.2% of hash power[xxvii], it is likely that they contribute about half of that power or more. Throw in the other chipfabs in proportion to the sales figures mentioned above, as well as any chipfabs that don’t sell to the public, and we will assume that chipfabs provide at least 35% of direct hash power for this study.
Due to the laws of perfect competition discussed earlier, it can be assumed that only the most profitable miners are switched on at any given time, and that when a new generation of mining equipment is released, equilibrium is reached very quickly where all miners are operating at a similar cost basis.
References
[i] Bitmain “Antminer S9i-14 TH/s”, https://shop.bitmain.com/?lang=en, (accessed 31 July 2018)
[ii] Canaan Creative, 2018, “Avalon 8 Series (Qty 40)- Shipping now Best value” http://archive.is/SkBgK
[iii] Hruska, J., 2013. “Intel’s former chief architect: Moore’s law will be dead within a decade” http://archive.is/9WNf8
[iv] Koomey, J. et al., 2010. Implications of Historical Trends in the Electrical Efficiency of Computing. IEEE — Annals of the History of Computing, 33(3), pp. 46–54.
[v] Johnson, P., Marker, T., 2009, “Data Centre Energy Efficiency Product Profile”, http://archive.li/ng9MO
[vi] Song, Z., Zhang, X., Eriksson, C., 2015. “Data Center Energy and Cost Saving Evaluation” http://archive.is/sEH60
[vii] Ni, J., Bai, X., 2016, “A review of air conditioning energy performance in data centers”, Renewable and Sustainable Energy Reviews, Volume 67 https://www.researchgate.net/publication/308343722_A_review_of_air_conditioning_energy_performance_in_data_centers (accessed 22 June 2018)
[viii] ABC News, 2018, “Iceland will soon use more energy mining bitcoins than powering its homes”, http://archive.is/qW31p
[ix] Malkin, S., 2018, “The Cheapest (and Best) Places for Bitcoin Mining”, https://cryptocurrencynews.com/daily-news/mining/cheapest-places-mining-bitcoin/ (accessed 22 June 2018)
[x] Blockchain.info,2018, “Total Transaction Fees”, http://archive.is/ct1ZY
[xi] International Energy Agency, 2017. “Key World Energy Statistics” http://archive.is/gYYIw
[xii] Sovacool, B. K., 2008. “Valuing the greenhouse gas emissions from nuclear power: A critical survey.” Economic Policy, Volume 36, p. 2950. http://archive.is/EaI5W
[xiii] Moomaw, W. et al., 2011. Annex II: Methodology. In IPCC: Special Report on Renewable Energy Sources and Climate Change Mitigation, Geneva: IPCC.
[xiv] Lazard, 2017. “Levelized Cost of Energy 2017” http://archive.is/UwOJA
[xv] Asciento, R. & Lawrence, A., 2013. Will energy prices power US datacenter growth or short-circuit energy efficiency? http://archive.is/RFrKU
[xvi] Meyer, D. 2018, “Mining a Bitcoin Costs About as Much as Buying One These Days” http://archive.is/flVWv
[xvii] Huang, Z. 2018, “This could be the beginning of the end of China’s dominance in bitcoin mining” http://archive.is/DWBla
[xviii] Clenfield, J. & Alpeyev, P., 2014. “The Other Bitcoin Power Struggle” http://archive.li/n8qT2
[xix] CSImarket.com, 2018, “Semiconductors Industry Profitability”, http://archive.is/M0XV8
[xx] CSImarket.com, 2018, “Computer Hardware Industry Profitability”, http://archive.is/eqTiw
[xxi] Genesis Mining, 2018, “Homepage”, https://www.genesis-mining.com/ (accessed 14 June 2018)
[xxii] Genesis Mining, 2018, “Pricing”, https://www.genesis-mining.com/pricing (accessed 14 June 2018)
[xxiii] Genesis Mining, 2018, “Customer Service”, https://www.genesis-mining.com/customer-service (accessed 14 June 2018)
[xxiv] Malwa, S. 2018, “Bitmain Considers Billion Dollar IPO after Expansion in Other Sectors” http://archive.li/IzFkq
[xxv] Schmidt, B. 2018, “Crypto’s 32-Year-Old Billionaire Mining King Is Mulling an IPO” https://archive.li/v5uXk
[xxvi] Cheng, E. 2018, “Secretive Chinese bitcoin mining company may have made as much money as Nvidia last year” http://archive.is/mm1qg
[xxvii]Blockchain.info, “Bitcoin Hash rate Distribution”, http://archive.fo/jPsCb
[xxviii]Bitmain, IPO Investor’s Deck. Original Chinese: https://www.docdroid.net/NhwMGIH/f.pdf#page=20
English Translated: https://drive.google.com/file/d/11IweEa_80ol4EjPmoRdIYTiBqYd_-y2T/view
