PDF hosted at the Radboud Repository of the Radboud University

Nijmegen

The following full text is a publisher's version.

For additional information about this publication click this link.

http://hdl.handle.net/2066/179131

Please be advised that this information was generated on 2024-04-07 and may be subject to
change.
 Available online at www.sciencedirect.com

ScienceDirect

Sustainability of bitcoin and blockchains

Harald Vranken1,2
Bitcoin is an electronic currency that has become increasingly validation of transactions, and can provide lower transac-
popular since its introduction in 2008. Transactions in the tion fees for payments [1,2]. The Bitcoin Foundation
bitcoin system are stored in a public transaction ledger (‘the provides some centralized governance for standardiza-
blockchain’), which is stored in a decentralized, peer-to-peer tion, protection and promotion of bitcoin, but it does
network. Bitcoin provides decentralized currency issuance and not act as a central bank and does not issue currency [3].
transaction clearance. The security of the blockchain depends
on a compute-intensive algorithm for bitcoin mining, which Bitcoin was introduced in 2008 by Satoshi Nakamoto [4],
prevents double spending of bitcoins and tampering with which is a pseudonym of an author or group of authors
confirmed transactions. This ‘proof-of-work’ algorithm is whose identity is covered in mysteries. The term ‘Bitcoin’
energy demanding. How much energy is actually consumed, is often refers to the system, while the term ‘bitcoin’ or
subject of debate. We argue that this energy consumption BTC refers to the unit of currency. In this paper, for
currently is in the range of 100–500 MW. We discuss the simplicity we just use the term bitcoin. Bitcoin is an
developments in bitcoin mining hardware. We also briefly electronic, virtual currency that has no physical represen-
outline alternative schemes that are less energy demanding. tation such as coins or banknotes. The bitcoin ecosystem
We finally look at other blockchain applications, and argue that is a network of users that communicate with each other
also here energy consumption is not of primary concern. using the bitcoin protocol via the Internet. The bitcoin
protocol is available as an open source software applica-
tion and allows users to store and transfer bitcoins for
Addresses
1
Open University of the Netherlands, P.O. Box 2960, 6401 DL Heerlen,
purchasing and selling goods, or to exchange bitcoins for
The Netherlands other currencies. The issuance of bitcoins takes places in
2
Radboud University, P.O. Box 9010, 6500 GL Nijmegen, The the network while handling transactions in a process
Netherlands called bitcoin mining. The bitcoin network started in
2009 and ever since bitcoin has been the most popular
Corresponding author: Vranken, Harald ([email protected])
decentralized currency. In January 2017 there were
16 million bitcoins in circulation with a total value of
Current Opinion in Environmental Sustainability 2017, 28:1–9 roughly 16 billion US dollars, although the exchange rate
This review comes from a themed issue on Sustainability governance of bitcoins has shown very large fluctuations.
Edited by Carolien Kroeze, Marjolein Caniels, Dave Huitema and
Harald Vranken Both scientific and professional literature on digital cur-
For a complete overview see the Issue and the Editorial
rencies, with bitcoin as prime example, is extensive.
Some provide gentle, general introductions to the tech-
Available online 29th May 2017
nology applied in bitcoin (e.g. [5]), while others provide
Received: 10 February 2017; Accepted: 23 April 2017 more detailed overviews of the technical operation of
http://dx.doi.org/10.1016/j.cosust.2017.04.011 bitcoin (e.g. [6
,7
,8]) as well as economical and financial
1877-3435/ã 2017 Elsevier B.V. All rights reserved. aspects (e.g. [9
]).

In this review paper we provide an overview and synthesis

of recent literature published in the last two years that
addresses the sustainability of bitcoin. The sustainability
of bitcoin is depending on a mix of environmental
Introduction [10

,11

], economical [1,12], financial [2,13,14] and eth-
People have been using currencies for thousands of years. ical [15] aspects. Bitcoin may pose risks to the stability of
Initially, currencies were minted directly from precious the current financial system, while also lack of controls
metals such as gold and silver. Later on, paper money was over bitcoin exchanges and the volatility of the bitcoin
introduced and the face value of cash was decoupled from currency raises concerns. Our focus in this review is on
its nominal value, but currencies were still backed up by sustainability in the context of environmental and eco-
gold depositories. Nowadays, fiat currencies are allowed nomical aspects. We try to answer the question whether
to float freely, only backed up by the faith and credit of the bitcoin system is sustainable given the energy con-
the states that issue them. Bitcoin is a decentralized sumption required for bitcoin mining, which has been
system that attempts to overcome the weaknesses of fiat subject of debate in the last few years. The contributions
and gold-based currencies. It is not governed by central of this paper are: firstly, to synthesize and critically assess
authorities, such as governments or central banks, and the viewpoints in scientific literature; and finally, to argue
intermediaries for currency issuance or settlement and that the energy consumption of the bitcoin system is not

www.sciencedirect.com Current Opinion in Environmental Sustainability 2017, 28:1–9

2 Sustainability governance

excessive, which stands in contrast to the public opinion mining process for the current block and starts mining for
that bitcoin mining is a gross waste of energy. We explore a new block. The node that won the block receives a
four subquestions: What factors play a role in the energy block reward, which is a fixed amount of new bitcoins.
consumption of bitcoin mining, how large is this energy Hence, the issuance of bitcoins (minting) is done during
consumption, does this impede sustainability, and if so are the bitcoin mining process. The node that won the block
there alternatives that can reduce energy consumption? In also receives the transaction fees for every transaction
the following sections we outline the basic operation of the included in the block. Every 10 min on average, a node is
bitcoin system, we summarize trends in the hardware used able to mine a new block. It can be the case that multiple
for bitcoin mining, we discuss the energy footprint of nodes simultaneously generate a valid block, which
bitcoin mining, we present some of the alternatives that causes that multiple versions of the blockchain (‘forks’)
have been proposed to reduce energy consumption, and we occur temporarily. Forks are resolved as soon as one of the
briefly discuss other applications of the blockchain tech- forks contains more blocks. The computations to find and
nology that is at the basis of the bitcoin system. verify a cryptographic hash of a block during bitcoin
mining allows the bitcoin network to gain consensus
Overview of the bitcoin system about the state of transactions. This elegantly solves
The bitcoin system is a distributed, peer-to-peer network. the issue of double spending and hence an amount of
There is no central server or point of control, and all nodes bitcoins cannot be spent twice. The bitcoin mining
in the network are equal peers. Each transaction to process decentralizes the currency issuance and the trans-
transfer an amount of bitcoins among users is transmitted action clearing normally done by central banks and clear-
to the bitcoin network where it is stored in a distributed inghouses. In economics bitcoin is considered as money
transaction ledger, the blockchain. The blockchain con- to some extent, since it offers a unit of account, means of
tains the entire history of bitcoin transactions. Each node payment, and store of value [1,3]. It can even be argued
in the network stores a (complete or partial) copy of the that bitcoin has an intrinsic value due to the computa-
blockchain. New transactions are propagated rapidly tional effort for bitcoin mining [17].
across the nodes in the network. A transaction is in fact
a transfer from a source of funds (called an input) to a Each block does not only contain transactions, but also
destination (called an output). Transaction inputs and the hash of the previously accepted block in the block-
outputs are not related to accounts or balances: an input is chain. Hence, the blocks in the blockchain are linked to
a reference to an unspent transaction output of the sender each other: they form a chain of blocks, thence the term
in a previous transaction. Before forwarding a transaction ‘blockchain’. This provides security, as a node with
to its neighbors, each node first verifies the transaction, malicious intent cannot easily replace or modify an
which includes checking the syntax and structure, and already accepted transaction or add a new transaction
whether it is a valid transfer of an amount of yet unspent to an already accepted block, since this would require
transaction outputs. Each node independently verifies to redo the computations to find a valid hash for the
the transactions received, propagates valid transactions, modified block. And since new blocks are continuously
and builds a pool of valid transactions. The valid transac- added to the blockchain, each block linking to the previ-
tions are added to the blockchain in a process called ous block, also the hashes of the newly added blocks
bitcoin mining. Each node collects a number of valid would have to be recomputed.
transactions into a block and tries to compute a crypto-
graphic hash of the block that meets certain constraints The initial block reward was set to 50 BTC. The reward is
(based on the ideas of Hashcash [16]). A cryptographic halved every 210 000 blocks, which is approximately
hash is a kind of checksum for the block, that is one-way every four years. This will continue until 2140 when
(meaning that it is easy to compute a hash of a given the mining reward drops below 108 BTC, which is
block, but difficult to compute a block that matches a the minimal unit of bitcoin also known as satoshi. After-
given hash) and collision resistant (meaning that it is wards, transaction fees will provide the necessary incen-
difficult to find two blocks that yield the same hash). tive to continue mining of new blocks [18]. The bitcoin
Finding a hash that meets the constraints imposed by the protocol includes an algorithm to regulate that on average
bitcoin system, is a compute-intensive task that can be every 10 min a new block is mined, by adjusting the
executed only by brute-force trying. This implies a race difficulty to find a valid hash. This is required to keep up
among the nodes in the network to find a valid hash as with the improvements in the performance of mining
quickly as possible. The first node that finds such a hash, hardware which allows bitcoin miners to compute more
wins the block, which means that this block is added to and more hashes per second.
the blockchain and propagated to the network. Although
computing a valid hash is difficult, verifying whether a Hardware for bitcoin mining
hash is valid is easy and hence each node that receives the Bitcoin mining is attractive since it offers a strong finan-
block can quickly identify whether the new block is valid. cial incentive. For each block mined, the miner receives a
When a node receives a new valid block, it stops the block reward as well as the transaction fees of the

Current Opinion in Environmental Sustainability 2017, 28:1–9 www.sciencedirect.com

 Sustainability of bitcoin and blockchains Vranken 3

transactions in the block. As bitcoin gained in popularity, (ASICs) containing dedicated circuitry that is optimized
an arms race started among miners. Bitcoin miners ini- to perform hashing computations as efficiently as possi-
tially used general-purpose computers, but they quickly ble. Butterfly Labs, ASICMiner and Avalon were the first
switched to more dedicated hardware that offered higher companies that provided ASICs for bitcoin mining,
performance (in terms of hash rate R, measured in the financed by online presales. ASICMiner initially did
number of hashes (h) computed per second) at lower not ship ASICs to customers, but ran the ASICs in their
energy costs (in terms of energy efficiency E, measured own data center, which allowed them to capture a large
in the number of hashes computed per Joule). This fraction of the total network hash rate. These first ASIC
dedicated hardware for bitcoin mining has developed manufacturers were very successful. Other companies
in a remarkable way, and bitcoin miners even self- with greater capitalization quickly followed and devel-
financed hardware and software development [19

,20]. oped the next generations of ASICs with improved tech-
nology. Currently, the most advanced technologies are
The bitcoin mining hardware has seen four generations only utilized by chip manufacturers that run their ASICs
[19

,20,17], see Table 1. Initially miners used general- in their own data centers located in areas that have low-
purpose computers, in which the actual computations are cost energy and cooling. The bitcoin mining industry is
performed by the Central Processing Unit (CPU). however very competitive. For instance, the Swedish
Although modern CPUs can execute software with a company KnCminer operated data centers located in
certain amount of parallelism, and multiple threads can the Arctic circle to benefit from locally sourced hydro-
be executed in parallel on multicore CPUs, they are not electric power and cool air at extremely low cost, but still
optimized for bitcoin mining. This first generation of went bankrupt mid 2016. Many large miners are located
bitcoin mining hardware using CPUs, is the least power- near cheap sources of electricity, such as hydroelectric
ful and the least energy efficient. As the difficulty of dams (China, Republic of Georgia) and geothermal power
mining increased, the operational costs of CPUs plants (Iceland).
exceeded the profits from mining.
Figure 1 shows the daily revenue in US dollars per Gh/s
The second generation occurred at the end of 2010 when earned by bitcoin miners in the period 2011–2016. The
bitcoin miners started to use the Graphics Processing figure combines historical data on the mining revenue (i.
Unit (GPU) in the graphics cards of their computers. e. block rewards and transaction fees) and the hash rate of
These GPUs are designed to perform complex graphics the bitcoin network with US dollar to BTC exchange rate.
calculations with lots of parallelism, which can be used The drops at the end of 2012 and mid 2016 correspond to
efficiently for bitcoin mining. GPUs offered higher hash the transition of the block reward from 50 BTC to 25 BTC
rates and better energy efficiency than CPUs. and from 25 BTC to 12.5 BTC. The horizontal lines show
the estimated daily energy cost per Gh/s for CPUs, GPUs,
As the use of GPUs became more widespread, bitcoin and a number of FPGAs and ASICs, including five gen-
miners started to look for more powerful and more effi- erations of ASICs in Bitmain’s Antminer product line.
cient alternatives. The third generation occurred mid Bitmain Technologies, founded in 2013, is currently one
2011 when miners switched to Field Programmable Gate of the leading ASIC manufacturers that ship ASICs to
Arrays (FPGAs). The circuits in an FPGA can be config- customers. When the revenue of mining drops below
ured and programmed by users after manufacturing. these costs, profit turns negative and miners have to
Bitcoin miners customized FPGAs to support mining, switch to more efficient hardware [21]. Note that this
which allowed to increase hash rates even further at lower figure is in line with the analysis by Taylor [19

] and
power consumption. The popularity of FPGAs was brief, shows costs for hardware that can be purchased by private
since the fourth generation appeared quickly. customers and run at electricity costs of 200 USD/MWh.
Electricity costs however vary widely in different coun-
The fourth generation appeared early 2013 with the tries, and even within countries, depending on infrastruc-
introduction of Application-Specific Integrated Circuits ture and geography. For instance, in 2015 the electricity
prices in OECD countries ranged for consumers from
Table 1 75.33 USD/MWh in Mexico to 337.38 USD/MWh in
Hash rate and energy efficiency (orders of magnitude) of four Denmark, and for industry from 35.34 USD/MWh in
generations of bitcoin mining hardware (data source: en.bitcoin. Norway to 263.33 USD/MWh in Italy (source: Interna-
it/wiki/mining_hardware_comparison) tional Energy Agency, www.iea.org). Industrial users run
Hardware Introduction Hash rate (h/s) Energy efficiency (h/J) purpose-built data centers comprised of specialized ser-
CPU 2009 105–10 8 104–10 5
vers that integrate arrays of ASICs (‘ASIC clouds’) offer-
GPU Late 2010 106–10 9 105–10 6 ing better performance and energy efficiency [20].
FPGA Mid 2011 108–10 10 10 7
ASIC Early 2013 1010–10 13 108–10 10 Bitcoin miners did not only participate in grass-root
efforts to produce efficient hardware, they also cooperate

www.sciencedirect.com Current Opinion in Environmental Sustainability 2017, 28:1–9

4 Sustainability governance

Figure 1

103
Daily bitcoin mining revenue and energy costs (USD per Gh/s)

102
CPU

101

GPU

0
10
FPGA (Bitforce SHA 256)

FPGA (Buterfly Labs Mini Rig)

10–1
ASIC (Avalon batch 1, 110 nm)

10–2 ASIC (AntMiner S1, 55 nm)

ASIC (AntMiner S4, 28 nm)

10 –3 ASIC (AntMiner S7, 28 nm)

ASIC (AntMiner S9, 16 nm)

10–4
2011-1-1 2012-1-1 2013-1-1 2014-1-1 2015-1-1 2016-1-1 2017-1-1
Date
Current Opinion in Environmental Sustainability

Daily mining revenue and daily mining energy cost for different types of hardware (USD per Gh/s) (data sources: www.blockchain.info for daily
mining revenue; en.bitcoin.it/wiki/mining_hardware_comparison for energy costs).

in mining pools in which participants split up the com- by a factor S3 per process generation. To maintain the
putations to mine a block. If a block is mined, each same power usage, the transistor energy efficiency also
participant is rewarded according to their contribution. has to improve with a factor S3. This is achieved by scaling
the transistor capacitance, which improves energy effi-
The bitcoin arms race increases the capital expenditure, ciency by a factor S, and by scaling the threshold and
which throws up barriers for newcomers to enter and operating voltages, which provides another factor S2
causes miners that cannot keep up to drop out. This improvement in energy efficiency. However, Dennard’s
leads to an oligopolistic market. According to data from scaling no longer holds for process generations below
bitcoinchain.com, the five largest miners, which are 90 nm, since further scaling of the threshold voltage
mostly based in China, mined over 85% of the blocks causes unacceptable levels of current leakage, and there-
in 2016. This implies several risks, such as government fore the operating voltage has to remain roughly constant.
interventions and undermining bitcoin’s principle of a Instead of improving the energy efficiency by S3, in post-
decentralized currency. Dennard scaling the energy efficiency can only be
improved by S. Hence, with each process generation
An interesting question is how bitcoin mining ASICs will we face a shortfall of S2. While transistor count continues
evolve in the near future. The semiconductor industry to increase according to Moore’s law, the per-transistor
has been introducing new CMOS process technology speed and energy efficiency improvements slow down
generations at a fairly constant two-year pace [22]. With exponentially [23,24]. To deal with this, more and more
each new generation, the dimensions of transistors on portions of chips are not used all the time, or not at full
pffiffiffi down further by a factor S, which
chips are scaled frequency (which is referred to as ‘dark silicon’) [25]. This
typically is 2. According to Dennard’s classic scaling caused a shift to multicore design in 2005. Some applica-
theory, by scaling the dimensions (and consequently the tions can benefit from specialized, heterogeneous cores
electrical characteristics) of a transistor with a factor S, the that can be dynamically powered up for a given workload
transistor count increases by a factor S2 (Moore’s law) and as in servers [26], or energy-efficient cores for computa-
the transistor switching frequency increases by a factor S, tionally intensive applications [23]. However, this is not
while keeping chip area and chip energy usage the same. the case for bitcoin mining ASICs that continuously
Hence, the computational capabilities of chips increase operate at peak performance, which results in extremely

Current Opinion in Environmental Sustainability 2017, 28:1–9 www.sciencedirect.com

 Sustainability of bitcoin and blockchains Vranken 5

high and sustained power consumption [27]. Since 2256  1. Hence, the only way to find a valid hash is to
2005 also the search has initiated for new types of randomly try nonce values. This scheme is called ‘proof-
switches that improve performance and energy efficiency. of-work’. The bitcoin network controls the difficulty for
It is unlikely that a new switch faster than CMOS tran- finding a valid hash by adjusting the target T every
sistors and consuming less power will be introduced into 2016 blocks, with the aim of keeping the average time
manufacturing on short term. Carbon nanotubes are to mine a new block near 10 min. The target is expressed
promising, but it still will require several years until this in terms of the difficulty D by D = Tmax/T, where Tmax is
reaches the manufacturing stage. 3D power scaling tech- the largest possible value of the target (which is (216  1)
nology allows the continuation of Moore’s law for the next 2208  2224). The probability that a nonce value yields a
10–15 years via power-efficient vertical transistors. Even- valid hash therefore is p = T/2256 = Tmax/(2256D)  1/
tually, switches will reach a fundamental performance (232D). The number of trials for choosing a nonce value
limit, and any further improvement in computing perfor- that yields a valid hash is approximately geometrically
mance can solely come from innovatives in system design distributed (assuming that these trials are independent,
[22]. The SHA-256 algorithm used for computing the which of course does not hold for a single miner, but
block hashes in bitcoin mining however does not lend considers that multiple miners are independently per-
itself to significant micro-architectural design modifica- forming the computations simultaneously). At rate R the
tions. The only improvement for bitcoin mining ASICs is expected time to find a valid nonce value therefore is
to migrate to the latest process technologies and possibly 1/( pR) = 232D/R, which equals 600 s. The hash rate of the
apply custom library cells or even custom physical layout bitcoin network then is R  232D/600. Combining this
[27]. Hence, the future improvement in performance and with the energy efficiency E, the estimated power con-
energy efficiency of bitcoin mining ASICs is expected to sumption of the bitcoin network is P = R/E  232D/
slow down. (600E) [10

].

Figure 2 shows the power consumption (orders of magni-

Energy costs of bitcoin mining tude) for various bitcoin mining hardware. Obviously, for
Next to the capital expenditure for bitcoin mining hard- any given date the estimated power consumption is
ware, the main costs for bitcoin mining are the operational realistic only when considering the hardware available
costs for running the hardware, which are mainly energy at that time (see Table 1). The figure indicates that in
costs. There has been lot of debate on the total energy January 2017 the actual power consumption could vary
consumption of bitcoin mining, not only on Internet fora from 45 MW (using state-of-the art ASICs with
but also in scientific literature [10

,11

,20]. The esti- 5  1010 h/J energy efficiency) up to 450 TW (using early
mates vary considerably, ranging from an energy con- generations of CPUs with 5  103 h/J energy efficiency).
sumption that is equivalent to the electricity generated by Since the worldwide annual electricity consumption is
a small power plant (in the order of 10 MW) up to the about 2.3 TW, it is clear that 450 TW is completely
electricity consumption of small to medium-sized coun- unrealistic. A more realistic upper bound on the energy
tries such as Denmark, Ireland or Bangladesh (in the consumed can be derived when assuming that the reve-
range of 3–6 GW). nue of bitcoin mining (see Figure 1) would be totally
spent on energy costs (hence ignoring capital expendi-
O’Dwyer and Malone analyzed the energy footprint of ture). The daily revenue of bitcoin mining, including
bitcoin mining in 2014 [10

]. Their analysis is based on block rewards and transaction fees, on January 1,
the observation that the power consumption of the bitcoin 2017 was 1 961 203 USD (according to www.
network (P, measured in W) can be computed from the blockchain.info). This is a plausible number when con-
hash rate of the bitcoin network (R, measured in h/s) and sidering that one block is mined every 10 min, which
the energy efficiency of the bitcoin mining hardware (E, yields a daily revenue of 1 800 000 USD (at a block
measured in h/J): P = R/E. reward of 12.5 BTC and an exchange rate of
1000 USD) not considering transaction fees. With this
During the mining process, the miner computes the hash revenue, the upper bound on the energy consumption is
of a block of transactions. A block also contains other data, in the range of 400 MW (electricity price of 200 USD/
such as the hash of the latest accepted block in the MWh) up to 2.3 GW (electricity price of 35 USD/MWh).
blockchain, and a ‘nonce’ value that the miner can choose When taking 60 USD/MWh as an average case for elec-
randomly. The aim of the miner is to find a nonce value tricity price, the energy consumption is 1.3 GW. The
such that the hash of the block is smaller than a target corresponding energy efficiency then is 1.8  109 h/J,
value T. In the bitcoin network, the 256-bits crypto- and hence it is clear that bitcoin mining currently is only
graphic hash of a block B is computed by applying the profitable when applying ASICs.
SHA-256 hash function [28] twice, h(B) = SHA256
(SHA256(B)), which yields a hash that behaves approxi- An even more accurate estimation of the energy con-
mately as a uniformly random value between 0 and sumption is derived when also considering the capital

www.sciencedirect.com Current Opinion in Environmental Sustainability 2017, 28:1–9

6 Sustainability governance

Figure 2

1015

1014 CPU
13
10
GPU
1012

11 FPGA
10
Power usage of bitcoin network (W)

10
10

9
10
ASIC
108

107

6
10

105

104

103

102

101

100
2011-1-1 2012-1-1 2013-1-1 2014-1-1 2015-1-1 2016-1-1 2017-1-1
Date
Current Opinion in Environmental Sustainability

Estimated power usage of bitcoin network (232D/(600E)) for various hardware (energy efficiency ranges (orders of magnitude) according to Table 1
(data sources: www.blockchain.info for historical data on difficulty; en.bitcoin.it/wiki/mining_hardware_comparison for energy efficiency of
hardware).

expenditure. Magaki et al. explored the design of pur- ASICs from older process technologies running at low
pose-built data centers running servers with large arrays of power [27]. Bitcoin mining is very competitive. Bitcoins
ASICs (‘ASIC clouds’) dedicated to bitcoin mining [20]. will be mined by those who can do it most cheaply, and
They consider three designs in which either energy, costs others will be put out of business. It is therefore likely that
or total cost of ownership (TCO) are optimized, at an surviving miners run the latest hardware at locations
electricity price of 60 USD/MWh. In these three cases, offering the lowest electricity costs to be competitive
the electricity costs are 7.5%, 16.9%, and 13.7% of the and to maximize profit.
TCO. In the break-even case, where revenue equals
TCO, the energy consumption is 100, 230 and Estimates published in scientific literature vary
190 MW. The corresponding energy efficiency then is considerably:
in the range of 1.1  1010 to 2.4  1010 h/J.

O’Dwyer and Malone estimated that the total power
The ASICs that are currently being used by bitcoin consumption for bitcoin mining would be around
miners, are most likely a mix of the newest available 100 MW to 10 GW [10

]. Without further substantia-
and some older ASICs. The actual mix used in practice is tion, they conclude that an average of 3 GW would be
unknown. Bitcoin miners will not switch to newer hard- most plausible (which is comparable to the Irish
ware as long as mining with their current hardware is still national energy consumption). Our analysis however
profitable and the break-even point has not been reached shows that is overestimated.
yet at which revenues have covered the capital and
McCook argues that chip-fabricator miners, who apply
operational expenditure of their current hardware. The the ASICs that they design and manufacture them-
future trend may well be to apply massive amounts of selves for mining, can mine for up to 30% cheaper than

Current Opinion in Environmental Sustainability 2017, 28:1–9 www.sciencedirect.com

 Sustainability of bitcoin and blockchains Vranken 7

retail miners, and that they form the vast majority of the their possession to themselves and add a predefined
hash power [11

]. Applying the 80-20 rule, assuming percentage as reward. In the mining process, still the
chip fabricators hold 80% and retail miners hold 20% of hash of a block has to be computed that is smaller than a
the hash power, the energy efficiency on average is target value. A block however does not include a nonce
estimated at 2.5 Gh/J, which corresponds to a power value that can be modified by the miner, but a time-stamp
consumption of 120 MW. that changes every second. Hence, miners cannot rely on

Magaki et al. state that the global power budget dedi- computational power, but they can only compute one
cated to ASIC clouds is estimated by experts to be in hash every second. The miner that wins the block,
the range of 300–500 MW [20]. receives the transaction reward. The difficulty is deter-
mined individually for every user: it is inversely propor-
We conclude that although the energy consumption could tional to the coin age, which is the amount of coins times
be as low as 45 MW when solely using the latest bitcoin the time period that the user held these coins. Hence,
mining ASICs, in practice the energy consumption most users with a large coin age have a higher chance to mine a
likely is in the range of 100–500 MW (which corresponds block. When a block is mined that includes a coinstake
to 3–16 PJ per year). Hence, the order of magnitude of the transaction, the coin age of the winner is reset. Hence,
energy consumption is 100 MW. proof-of-stake is a raffle-like scheme, with repeatedly
occurring new chances for all participants [36–38]. Also
To put things into perspective, McCook also compares a combination of proof-of-work and proof-of-stake has
the sustainability of bitcoin mining with the sustainability been proposed, in which a fraction of the proof-of-work
of gold mining and the banking system [11

]. The energy block reward is raffled among all active nodes, while their
used per year for gold mining and recycling is estimated at stake determines the amount of raffle tickets [39].
500 PJ, for printing paper banknotes and minting coins at
40 PJ, and for the banking system, considering ATMs and Another alternative is proof-of-space, where the miner
bank branches (which of course provide more services must employ a specified amount of memory to compute
than just handling transactions), at 2340 PJ. Compared to the proof [40,41]. In proof-of-space-time, the miner must
these numbers, the energy used for bitcoin mining in the prove that he stored data over a period of time [42].
range of 3–16 PJ is relatively small. Still, the proportion of
bitcoin in the current financial system is relatively small, Although these alternatives largely reduce the energy
and when bitcoin scales up, so will the effort for bitcoin consumption as with proof-of-work, there still are security
mining. issues when applying them to public blockchains [39,38].

Another line of thought to deal with the criticism that Blockchains

proof-of-work as applied in bitcoin wastes energy, is to Blockchain is at the basis of currencies such as bitcoin, but
replace the computation of hashes by more ‘meaningful’ it can also be used in many other financial and commercial
tasks. This has been applied in other electronic curren- applications [43–49,35]. A prominent example is smart
cies. For instance, NooShare proposes the scheduling of contracts, for instance as offered in Ethereum [50]. A
arbitrary Monte-Carlo simulations as a proof-of-work, contract can execute a transfer when certain events hap-
Primecoin proposes the computation of long chains of pen, such as payment of a security deposit, while the
prime numbers (Cunningham chains), and Permacoin correct execution is enforced by the consensus protocol
proposes proofs of retrievability [7
]. [51,52].

There are also other factors that impact the sustainability Blockchains can be classified as public blockchains, pri-
of bitcoin [29]. For instance, bitcoin is not suited for real- vate blockchains or consortium blockchains [35]. Bitcoin
time transactions due to the delay between the injection is an example of a public blockchain, in which all records
of a transaction into the bitcoin network and the inclusion are visible to the public and everyone can take part in the
of the transaction in a mined block that is added to the consensus process. A private blockchain is fully controlled
blockchain, and for the transaction actually to be con- by one organization, with a closed group of known parti-
firmed a sufficient amount of subsequent blocks has to be cipants, which implies a centralized rather than a decen-
added to the blockchain [30,31]. Other concerns are the tralized network. A consortium blockchain is partially
growing size of the blockchain, and security [32–34,7
]. decentralized, where transactions are validated by a
selected set of nodes. Private and consortium blockchains
Alternatives for proof-of-work may permission other users to read records in the block-
Various alternative consensus mechanisms have been chain. Public blockchains rely on a consensus protocol
proposed to address the energy consumption of proof- such as proof-of-work, which ensures that transactions
of-work [35]. In proof-of-stake, users are required to prove cannot be tampered as long an no single miner controls
the ownership of their amount of coins. Users create ‘ more than 50% of the network’s hash power. Transactions
coinstake’ transactions in which they send the coins in in private or consortium blockchains are editable as long

www.sciencedirect.com Current Opinion in Environmental Sustainability 2017, 28:1–9

8 Sustainability governance

as the major participants have reached an agreement, and 7. Tschorsch F, Scheuermann B: Bitcoin and beyond: a technical

survey on decentralized digital currencies. IEEE Commun Surv
hence a strong consensus protocol such as proof-of-work is Tutor 2016, 18:2084-2123 http://dx.doi.org/10.1109/
not required. This reduces security, but improves effi- COMST.2016.2535718.
This survey describes the basics of bitcoin, and discusses security
ciency and latency, and hence energy consumption is threats and privacy properties, as well as the proof-of-work scheme
barely an issue. and alternative approaches.
8. Garay J, Kiayias A, Leonardos N: The bitcoin backbone protocol:
analysis and applications. Annual International Conference on
Conclusion the Theory and Applications of Cryptographic Techniques.
In this review we described the basic operation of bitcoin Springer; 2015:281-310.
mining and we explored the developments in the hard- 9. Chuen DLK (Ed): Handbook of Digital Currency. Academic Press;
ware used for bitcoin mining. The proof-of-work scheme is
2015. ISBN: 978-0-12-802117-0.
This book covers technical, economical and financial aspects of bitcoin.
compute-intensive and hence energy demanding, but
essential for dealing with the double-spending problem 10. O’Dwyer KJ, Malone D: Bitcoin mining and its energy footprint.


Irish Signals & Systems Conference 2014 and 2014 China–Ireland
and security of the blockchain. The mining hardware has International Conference on Information and Communications
evolved from CPUs, GPUs and FPGAs to ASICs, with an Technologies (ISSC 2014/CIICT 2014). IET; 2014:280-285.
In this paper the energy footprint of the bitcoin network is computed. It is
exponential increase in performance and energy efficiency. estimated that the electricity consumption of bitcoin mining is on par with
It is expected however that this trend will slow down in the that of Ireland.
next decade. We discussed the energy footprint of bitcoin 11. McCook H: An Order-of-Magnitude Estimate of the Relative
mining, which has been subject of debate. Our estimates

Sustainability of the Bitcoin Network. 2015 In: https://www.
academia.edu/7666373/An_Order-of-Magnitude_Estimate_
show that the order of magnitude for the energy consump- of_the_Relative_Sustainability_of_the_Bitcoin_
tion is 100 MW. As bitcoin becomes more popular, the Network_-_3rd_Edition.
This paper compares the sustainability of the bitcoin network with the
effort for bitcoin mining will increase. Since bitcoin mining banking industry, the gold production industry, and the process of
is very competitive, only those miners will survive who printing and minting of physical currency.
apply the most competitive mining hardware and benefit 12. Harwick C: Cryptocurrency and the problem of intermediation.
from the lowest electricity costs. The sustainability of Independ Rev 2016, 20:569-588.
bitcoin on itself therefore is not primarily at risk due to 13. Grant G, Hogan R: Bitcoin: risks and controls. J Corp Account
energy consumption. We also briefly reviewed alternative Finance 2015, 26:29-35 http://dx.doi.org/10.1002/jcaf.22060.
schemes such as proof-of-stake, which are far less energy 14. Walch A: The bitcoin blockchain as financial market
demanding. Finally, we looked at other applications of infrastructure: a consideration of operational risk. N Y Univ J
Legisl Public Policy 2015, 18:837-893.
blockchain technology, which are currently receiving lots
of interest. Private and consortium blockchains are only 15. Angel JJ, McCabe D: The ethics of payments: paper, plastic, or
bitcoin? J Business Ethics 2015, 132:603-611.
partially decentralized, which relaxes the need and effort
16. Back A: Hashcash — A Denial of Service Counter-Measure. 2002
for proof-of-work schemes, and hence energy consumption In: ftp://sunsite.icm.edu.pl/site/replay.old/programs/hashcash/
may be barely an issue. hashcash.pdf.
17. Hayes AS: Cryptocurrency value formation: an empirical study
Acknowledgements leading to a cost of production model for valuing bitcoin.
Telemat Informat 2016 http://dx.doi.org/10.1016/j.
We kindly thank the anonymous reviewers for their valuable comments. tele.2016.05.005. (in press).
18. Kaskaloglu K: Near zero bitcoin transaction fees cannot last
References and recommended reading forever. The International Conference on Digital Security and
Papers of particular interest, published within the period of review, Forensics (DigitalSec2014); The Society of Digital Information and
have been highlighted as: Wireless Communication: 2014:91-99.

of special interest 19. Taylor MB: Bitcoin and the age of bespoke silicon. In


of outstanding interest

Proceedings of the 2013 International Conference on Compilers,
Architectures and Synthesis for Embedded Systems, CASES’13.
Piscataway, NJ, USA: IEEE Press; 2013. pp. 16:1–16:10. ISBN:
1. Ali R, Barrdear J, Clews R, Southgate J: The economics of digital 978-1-4799-1400-5. http://dl.acm.org/citation.cfm?id=2555729.
currencies. Bank Engl Q Bull 2014, 2014:276-286. 2555745.
2. Mikołajewicz-Wozniak A, Scheibe A: Virtual currency schemes This paper describes the history of bitcoin mining hardware, especially
— the future of financial services. Foresight 2015, 17:365-377 the development of ASICs, as well as the incentives and strategies of
http://dx.doi.org/10.1108/FS-04-2014-0021. bitcoin miners.

3. Beer C, Weber B: Bitcoin — the promise and limits of private 20. Magaki I, Khazraee M, Gutierrez LV, Taylor MB: ASIC clouds:
innovation in monetary and payment systems. Monet Policy specializing the datacenter. In Proceedings of the 43rd
Econ 2014, 4:53-66. International Symposium on Computer Architecture, ISCA’16.
Piscataway, NJ, USA: IEEE Press; 2016, 178-190 http://dx.doi.
4. Nakamoto S: Bitcoin: A Peer-to-Peer Electronic Cash System. org/10.1109/ISCA.2016.25:. ISBN: 978-1-4673-8947-1.
2008.
21. Wang L, Liu Y: Exploring miner evolution in bitcoin network. In
5. Zohar A: Bitcoin: under the hood. Commun ACM 2015, 58:104- Proceedings 16th International Conference on Passive and Active
113 http://dx.doi.org/10.1145/2701411. Network Measurement; vol 8995 of Lecture Notes in Computer
Science Series. Springer; 2015:290-302 http://dx.doi.org/
6. Antonopoulos AM: Mastering Bitcoin: Unlocking Digital 10.1007/978-3-319-15509-8_22.

Cryptocurrencies. O’Reilly Media; 2014. ISBN: 978-1-4493-7404-
4. http://chimera.labs.oreilly.com/books/1234000001802/index. 22. Gargini PA: How to successfully overcome inflection points, or
html. long live Moore’s law. Comput Sci Eng 2017, 19:51-62 http://dx.
This book outlines the technical operation of bitcoin. doi.org/10.1109/MCSE.2017.32.

Current Opinion in Environmental Sustainability 2017, 28:1–9 www.sciencedirect.com

 Sustainability of bitcoin and blockchains Vranken 9

23. Venkatesh G, Sampson J, Goulding N, Garcia S, Bryksin V, Lugo- 37. Bentov I, Gabizon A, Mizrahi A: Cryptocurrencies Without Proof of
Martinez J, Swanson S, Taylor MB: Conservation cores: Work. 2014arXiv:1406.5694.
reducing the energy of mature computations. In Proceedings of
the Fifteenth Edition of ASPLOS on Architectural Support for 38. Kiayias A, Konstantinou I, Russell A, David B, Oliynykov R: A
Programming Languages and Operating Systems. ACM; 2010 Provably Secure Proof-of-Stake Blockchain Protocol. Cryptology
http://dx.doi.org/10.1145/1736020.1736044. ISBN: 978-1-60558- ePrint Archive, Report 2016/889. 2016 In: http://eprint.iacr.org/
839-1; 205-218. 2016/889.

24. Esmaeilzadeh H, Blem E, Amant RS, Sankaralingam K, Burger D: 39. Bentov I, Lee C, Mizrahi A, Rosenfeld M: Proof of Activity:
Power challenges may end the multicore era. Commun ACM Extending Bitcoin’s Proof of Work Via Proof of Stake. Cryptology
2013, 56:93-102 http://dx.doi.org/10.1145/2408776.2408797. ePrint Archive, Report 2014/452. 2014 In: http://eprint.iacr.org/
2014/452.
25. Taylor MB: A landscape of the new dark silicon design regime.
IEEE Micro 2013, 33:8-19. 40. Ateniese G, Bonacina I, Faonio A, Galesi N: Proofs of space:
when space is of the essence. International Conference on
26. Hardavellas N, Ferdman M, Falsafi B, Ailamaki A: Toward dark Security and Cryptography for Networks. Springer; 2014:538-557.
silicon in servers. IEEE Micro 2011, 31:6-15.
41. Dziembowski S, Faust S, Kolmogorov V, Pietrzak K: Proofs of
27. Barkatullah J, Hanke T: Goldstrike 1: Cointerra’s first- space. Advances in Cryptology — CRYPTO 2015; vol 9216 of
generation cryptocurrency mining processor for bitcoin. IEEE Lecture Notes in Computer Science Series. Springer; 2015:585-
Micro 2015, 35:68-76. 605 http://dx.doi.org/10.1007/978-3-662-48000-7_29.

28. NIST: FIPS PUB 180-4, Secure Hash Standard (SHS). 2015 http:// 42. Moran T, Orlov I: Proofs of Space–Time and Rational Proofs of
dx.doi.org/10.6028/NIST.FIPS.180-4. Storage. Cryptology ePrint Archive, Report 2016/045. 2016 .
Report 2016/035 In: https://eprint.iacr.org/2016/035.
29. Extance A: Bitcoin and beyond. Nature 2015, 526:21-23.
43. Underwood S: Blockchain beyond bitcoin. Commun ACM 2016,
30. Decker C, Seidel J, Wattenhofer R: Bitcoin meets strong 59:15-17 http://dx.doi.org/10.1145/2994581.
consistency. In Proceedings of the 17th International Conference
on Distributed Computing and Networking, ICDCN’16. New York, 44. Umeh J: Blockchain double bubble or double trouble? ITNOW
NY, USA: ACM; 2016 http://dx.doi.org/10.1145/ 2016, 58:58-61.
2833312.2833321. 13:1–13:10. ISBN: 978-1-4503-4032-8.
45. Fanning K, Centers DP: Blockchain and its coming impact on

31. Karame GO, Androulaki E, Roeschlin M, Gervais A, Capkun S: financial services. J Corp Account Finance 2016, 27:53-57 http://
Misbehavior in bitcoin: a study of double-spending and dx.doi.org/10.1002/jcaf.22179.
accountability. ACM Trans Inform Syst Security 2015, 18:1-32 46. Yli-Huumo J, Ko D, Choi S, Park S, Smolander K: Where is current
http://dx.doi.org/10.1145/2732196. research on blockchain technology? A systematic review.
32. Gervais A, Ritzdorf H, Karame GO, Capkun S: Tampering with the PLOS ONE 2016, 11 http://dx.doi.org/10.1371/journal.
delivery of blocks and transactions in bitcoin. In Proceedings of pone.0163477.
the 22nd ACM SIGSAC Conference on Computer and 47. BitFury Group, Garzik J: Public Versus Private Blockchains — Part
Communications Security, CCS’15. New York, NY, USA: ACM; 1: Permissioned Blockchains. White Paper. 2015 In: http://bitfury.
2015, 692-705 http://dx.doi.org/10.1145/2810103.2813655:. com/content/5-white-papers-research/public-vs-private-pt1-1.
ISBN: 978-1-4503-3832-5. pdf.
33. Gervais A, Karame GO, Wüst K, Glykantzis V, Ritzdorf H, 48. BitFury Group, Garzik J: Public Versus Private Blockchains — Part
Capkun S: On the security and performance of proof of work 2: Permissionless Blockchains. White Paper. 2015 In: http://bitfury.
blockchains. In Proceedings of the 2016 ACM SIGSAC com/content/5-white-papers-research/public-vs-private-pt2-1.
Conference on Computer and Communications Security, CCS’16. pdf.
New York, NY, USA: ACM; 2016, 3-16 http://dx.doi.org/10.1145/
2976749.2978341:. ISBN: 978-1-4503-4139-4. 49. Forte P, Romano D, Schmid G: Beyond Bitcoin — Part I: A Critical
Look at Blockchain-Based Systems. Cryptology ePrint Archive,
34. Paul G, Sarkar P, Mukherjee S: Towards a more democratic Report 2015/1164. 2015 In: http://eprint.iacr.org/2015/1164.
mining in bitcoins. In In Proceedings of 10th International
Conference on Information Systems Security (ICISS); vol 8880 of 50. Wood G: Ethereum: A Secure Decentralised Generalised
Lecture Notes in Computer Science (LNCS). Edited by Prakash A, Transaction Ledger. Ethereum Project Yellow Paper. 2014:151 In:
Shyamasundar R. Proceedings of 10th International Conference http://gavwood.com/paper.pdf.
on Information Systems Security (ICISS); vol 8880 of Lecture Notes
in Computer Science (LNCS) Cham: Springer International 51. Luu L, Chu DH, Olickel H, Saxena P, Hobor A: Making smart
Publishing; 2014:185-203 http://dx.doi.org/10.1007/978-3-319- contracts smarter. In Proceedings of the 2016 ACM SIGSAC
13841-1_11. ISBN: 978-3-319-13841-1. Conference on Computer and Communications Security, CCS’16.
New York, NY, USA: ACM; 2016, 254-269 http://dx.doi.org/
35. Zheng Z, Xie S, Dai H, Chen X, Wang H: An overview of 10.1145/2976749.2978309:. ISBN: 978-1-4503-4139-4.
blockchain technology: architecture, consensus, and future
trends. Proceedings of 6th IEEE International Congress on Big 52. Luu L, Teutsch J, Kulkarni R, Saxena P: Demystifying incentives
Data 2017. in the consensus computer. In Proceedings of the 22nd ACM
SIGSAC Conference on Computer and Communications Security,
36. King S, Nadal S: PPcoin: Peer-to-Peer Crypto-Currency with CCS’15. New York, NY, USA: ACM; 2015, 706-719 http://dx.doi.
Proof-of-Stake. 2012. org/10.1145/2810103.2813659:. ISBN: 978-1-4503-3832-5.

www.sciencedirect.com Current Opinion in Environmental Sustainability 2017, 28:1–9