Bitcoin: Cryptopayments Energy Efficiency

Michel KHAZZAKA*
[email protected]
April 20, 2022

ABSTRACT

Bitcoin introduced a cryptographic peer-to-peer version of money that allows online

payments to be sent directly from one party to another without going through a
nancial institution. Many recent studies evaluated and criticised Bitcoin’s energy
consumption1 through its Proof of Work (PoW) consensus mechanism without
evaluating its ef ciency compared to classical electronic payment system.

Based on physics, information science and economics, we compute and compare the
energy consumption and de ne what is the energy ef ciency of both the current
monetary payment system and Bitcoin cryptopayment system. We demonstrate that
Bitcoin consumes 56 times less energy than the classical system, and that even at the
single transaction level, a PoW transaction proves to be 1 to 5 times more energy
ef cient. When Bitcoin Lightning layer is compared to Instant Payment scheme,
Bitcoin gains exponentially in scalability and ef ciency, proving to be up to a million
times more energy ef cient per transaction than Instant Payments.

INTRODUCTION convinced of the capabilities of the distributed ledger

technology4 (DLT) in payments, banking and
nance. The DNB paper did not compare Bitcoin
Bitcoin is designed and built to function as a world
PoW energy ef ciency with any parts of the classical
global currency and an online payment system. This
monetary and payments system. What is needed is a
is the promise declared in the 1st sentence of the
correct evaluation of Bitcoin functions and their
Bitcoin white paper abstract: “A purely peer-to-peer
energy consumption compared to their counterparts
version of electronic cash would allow online payments”.
in the classical electronic cash and payments systems.
While Bitcoin is still representing ≈ 42% of the total
Many have tackled this challenge without completing
cryptocurrencies’ market cap, many of its detractors
it. Changpeng Zhao the founder of Binance recently
continue their criticism of its Proof of Work
a s k e d f o r a ny d a t a a b o u t p a y m e n t s ’ v s.
consensus mechanism accusing it of being power-
cryptopayments energy consumption and Institut
hungry up to megalomania. The central bank of
Sapiens the French think “tech” recommended in its
Netherlands DNB compared its energy consumption
publication “Bitcoin, totem & tabou” for this work to be
to a whole country like Denmark or the Netherlands
performed: “banking industry, whose energy cost is
in the De Nederlandsche Bank paper “The carbon
considerable but never evaluated, recognized, nor published. It
footprint of bitcoin”2. Although most3 central banks do
would be interesting to calculate the energy cost of the banking
not recognise Bitcoin as legal tender, yet they are
sector…”

1 In march 2022, EU Markets in CryptoAssets regulation (MiCA) discussed banning completely proof of work based crypto-assets for claims
of energy inef ciency.
2 By Authors Juan Pablo Trespalacios and Justin Dijk, 2021

3 Central banks of Salvador and Central African Republic recognised Bitcoin as legal tender other countries have regulations that recognize

Bitcoin as analogue to foreign currencies such as Russia.

4 See our Cryptopayments Glossary issued in 2020 by France Payments Forum (in French) for the difference between blockchain and DLT as

well as the de nition of legal tender etc.

Bitcoin: Cryptopayments Energy Ef ciency 1/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
 The weaknesses in the previous studies are as follows: SCIENTIFIC APPROACH

Use of inaccurate numbers or incomplete

In order to study the energy consumption of Bitcoin
methodologies: for example the Cambridge
PoW cryptocurrency system and the classical
Bitcoin Electricity Consumption Index (CBECI)
electronic system, we start by setting the governing
which is based on a world average of electricity
mathematical and scienti c equations into play that
prices in USD and an average distribution of
will be used hereafter.
mining hardware leads to inaccurate results
varying between −50% (lower bound) up to
Work in physics, is the energy transferred to or from a
+120% (upper bound). This is a known limitation
system via the application of force along a
of the CBECI methodology in addition to its lack
displacement in spacetime6. In our case, a payment
of comparing 2 similar systems ef ciency. This
work is to transfer an amount of value money from a
paper will address both of these issues.
payer to a payee along a displacement over time. Note
Often partial or anti-bitcoin position: and that by nature an electronic transaction (noted Tx) can
usually do not account for both monetary systems travel the globe in near real-time so the notion of
and payments systems. For instance the central displacement in distance measured in kilometer has a
bank paper by the DNB “The carbon footprint of lower weight in the equation compared to the
bitcoin”, compares Bitcoin energy consumption to displacement in time which will be more weighing in
the debit card payment system alone basing their the equation. Here time is the time span δ t required
statements on the Cambridge index and the study to nish the payment work.
“The Energy Consumption of Blockchain Technology:
Beyond Myth” 5. Yet card payments are just an A payment work is to transfer an amount
intermediary step of the payment transaction, they of value called money from a payer to a payee [A]
mainly provide an authorisation of a transaction along a displacement over time.
and will at least require later inter-banking clearing
and settlement to become nal. On the other hand
The physical force7 applied is provided mainly
a bitcoin transaction is nal and covers end-to-end
through an electrical force causing a differential of
steps of the transaction, so the comparison is very
energy. From Newton's second law, it can be shown
incomplete.
that work on a free (no elds), rigid (no internal
degrees of freedom) body, is equal to the change in
It’s essential to compare Bitcoin energy consumption
kinetic energy Ek corresponding to the linear velocity
with all the aspects of the classical monetary
and angular velocity of that body W = ΔEk , where
payment system. This covers: banknotes and coins
the energy is the quantitative property that must be
cash management in ATM systems, card payments,
transferred to the system to perform work (and/or to
point of sale (POS) payments, banking and inter
heat it)8. It’s important to insist that energy is a
banking energy consumption etc. (see
conserved quantity; the law of conservation of
METHODOLOGY below)
energy states that energy can be converted in form,
We’ve endeavoured in our paper to answer but not created or destroyed. Energy like work is also
mathematically and scienti cally all these challenges measured in Joules in the international system9 but
for the bene t of decision makers, researchers, can also be measured in Wa t t × h o u r or kWh10
politicians, legislators and industry representatives. which is 1000Wh. Power is the amount of energy
transferred or converted per time unit:

5 The research used by DNB to criticise the PoW actually concludes by stating: “While their energy consumption is, indeed, massive, particularly when
compared to the number of transactions they can operate, we found that they do not pose a large threat to the climate, mainly because the energy consumption of PoW
blockchains does not increase substantially when they process more transactions”. But the central bank seemed to have missed this detail.
6 In its simplest form, it is often represented as the product of force and displacement.

7 Power and energy are scalar quantities in opposite to the vectorial form of the force.

8 Heat and mechanical work are special forms of a same value that is conserved and is called energy by Lord Kelvin and he called thermodynamics

the science that studies it.

9 One Joule is the energy transferred to an object by the work of moving it a distance of one metre against a force of one newton.

10 The conversion rule is 1kWh = 3 600 000 Joules or 1Wh = 3600J

Bitcoin: Cryptopayments Energy Ef ciency 2/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
fi
 E n erg y dW [0] The correct way to compare the classical monetary
Po w e r = = = F.v
Time dt and payments system to Bitcoin is to compare all
their common capabilities in terms of energy
Power is measured in Watts, joules per second, and consumption15. Bitcoin serves as a:
also in watt-hour/day or terawatt-hour per year
(TWh/yr) for example when quantifying the Monetary System: issuing, burning, and
electrical power of a data center or a country. For circulating the cryptocurrency bitcoin comparable
instance, the energy consumed by the biggest to a central bank with various commercial banks
datacenter in the world Equinix is 6.46 TWh/yr. The issuing and distributing central bank money and
whole digital services on the internet consume 2,000 commercial money.
TWh/yr. In France digital services consume 10
TWh/yr which is 0.5% of total digital energy. We Means of Payment allowing the transfer of the
will use the megawatt: MW or the version of TWh/ cryptoasset bitcoin from a payer to a payee. The
yr to measure power (power being an energy blockchain nodes serve as Payment Service
consumption here per year see [0]). We will also use Providers (PSP) similarly banks use card schemes,
kWh to measure energy.11 clearing and settlement mechanisms in the classical
electronic payments industry with central banks.
If Δ is the amount of work performed during a
The high level work breakdown structure of the
period of time of duration Δt, the average power Pavg monetary system can be simpli ed to the following
ΔW functions:
over that period is given by the formula: Pavg = .
Δt
Monetary mass issuing and circulation of
What follows is the evaluation of the total consumed
electronic money as well as the paper cash money
energy — that is power — of the monetary and
and coins,
payment systems worldwide compared to the energy
consumption of Bitcoin. Monetary distribution and lifecycle
management through the economy based on
METHODOLOGY supervised and regulated banking and nancial
institution service providers. This covers the
physical form of money distribution in secured
Money in the economy is a measure of work12, a value
vehicles, vaults, and ATMs as well as an electronic
commonly called price13. Today money is considered
form of money.
to be a nancial instrument issued by special monetary
authorities such as central banks. In this paper, we Bookkeeping services with central bank accounts
argue that money can be quali ed as a social in wholesale using central bank money, and
contract, in essence, and thus can be de ned as an customer banking service bookkeeping in retail
asset with an intrinsic power differential between two using core banking and online banking.
economic agents. Traditionally money serves three
functions: A unit of account as the foundation for Non-card payment services such as wire
economic metrology, a medium of exchange allowing to transfers, instalment automatic withdrawal like
transform a value into work through space in the form direct debit and other cross border or nancial
of payment transactions and a store of value messages operations using third party providers
transporting this value through time. It can be easily like Swift in addition to Clearing and Settlement
seen that there are natural relations between money, Mechanisms (CSM)
work14, energy and power.

11 To help understand the meaning of energy ef ciency and power vs energy, note that burning 1Kg of coal releases much more energy than
detonating 1Kg of TNT, but because the TNT reaction releases energy much more quickly, it delivers far more power than the coal.
12 Plato de ned it as a social convention while Aristotle as a measure of work See Aristotle's text here

13 Some time the price of money means the interest rate

14And also Proof of Work PoW

15It’s not in the scope of this paper to demonstrate that Bitcoin can serve as a currency and as a payment system. This is taken as a
hypothesis and as the promise of its Blockchain as stated by Satoshi’s white-paper.

Bitcoin: Cryptopayments Energy Ef ciency 3/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

𝒲
fi
fi
fi
fi
fi
fi
fi
fi
fi
 Card Payment services which include point of replaced by blockchain wallets. So they are out of
sales (POS) acceptance solutions and terminals, in scope for simpli cation reasons and to ensure an
addition to plastic card issuance and distribution, homogenous methodology.
vicinity and online payments authorisation through
a card scheme such as Visa, MasterCard or Carte Most of the studies16 available missed the point that
Bancaire locally in France. It also includes the it is a mistake to compare Bitcoin to Visa only since a
collection, clearing and settlement of those card scheme does not execute a payment transaction
transactions through clearinghouses. from end to end as Bitcoin does. A card scheme
ensures an authorisation in real time between the
It’s important to note that this paper is a global actors of the payments value chain: the bank of the
evaluation of payments worldwide, but in reality cardholder called the Issuer and the merchant’s
electronic payment systems are very fragmented and Acquirer bank. But most17 of the time the card
have different features and levels of ef ciencies in scheme and the two banks need to settle the
different regions and countries around the world. In transaction in a delayed step using central bank
addition, the carbon foot print in CO2 for instance is money and sometimes between corespondent banks
not a reliable approach, since many companies and and different central banks in case of a cross border
industries cover their carbon footprint by buying payment. In comparison, a transaction in bitcoin is
carbon credits. Their resulting carbon footprint is not nal in near real-time, it is a push payment in only
their real carbon emission. The only scienti cally one step and the nality time is set to be about 10
reliable approach is by the computation of the energy minutes on average (or much less through Lightning).
in terawatt-hour (TWh) per year required by each
system to function, and to compare their work Electronic System Crypto System
accomplished in this pure form of input energy
without taking into account the energy sources. Annual Issuing Included Included

Distribution ATM and POS N/A

SCOPE Bookkeeping Banks Blockchain

Card payments Acceptance, POS, N/A

The chosen scope is to compare the “Run” of both authorisation & CSM
Bitcoin and the existing electronic system: this means
Other payments Wiretransfer, debit, etc. Cryptopayment
to compare the energy consumption of the similar
running operations and to leave out of scope the Cheques Excluded N/A
“Build”of each system, such as the manufacturing of
ATM or miner units, the printing of banknotes and Finance services Excluded (insurance & Excluded (DeFi)
minting of coins. Although comparing the Build loans, etc.)

would have been bene cial for Bitcoin, it confers to it In scope of capabilities included for energy assessment
an unfair advantage. The issuance of bitcoins, called (highlighted) [T0]
mining is included since it is a part of the running
operations. The annual printing of banknotes and
Note that credit services are kept out of the scope of
minting of coins are included, the cash distribution
this study as well as DeFi nancial services and web
to ATMs and acceptance at electronic Point of Sales
3.0 use cases in order to focus only on Bitcoin vs
are also included to cater for the running of
monetary and payments industries. Electronic
electronic cash management. In addition all Payment
cheques payments are also kept out of scope for
Service Providers such as PayPal or any other
simpli cation considering their adoption decrease
marketplace payment or acceptance solution
(yet they still consume a considerable energy
providers are left out of scope since these can be
especially through printing, distribution and transit
proposing cryptopayments solutions also, or can be
back to bank).

16 For example DNB paper “The carbon footprint of bitcoin” or Arcane Research “The State of Lightning” — Oct 2021
17 There exist a card payment mode called “single mode” that allows the cardholder to be instantly debited and the merchant account to be
directly credited, but most of card payment transaction are in “dual mode” requiring to distinct steps: authorisation followed by collection of
translations at the end of day, then a request to a clearing and settlement mechanism for compensation with central bank money. This
distinction will not impact the energy consumption of the actors in general but only increase the speed of certain electronic payments
compared to cryptopayments.

Bitcoin: Cryptopayments Energy Ef ciency 4/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
fi
 An approach that compares the transacted values in ENERGY OF PRINTING & MINTING CASH
equivalent USD, not just transactions volumes is also
unsuitable. In both systems, energy consumption
Let’s start by calculating the energy consumption of
won’t be affected by transaction value. We have
the renewal rates of banknotes and coins. Note that
managed to exclude from this research 100% of this
we will not take into account the initial printing and
research any exchange value rates and transacted
minting of cash money according to our
amounts between Bitcoin and classical payments in
methodology of comparing only the run time energy
dollars or euros for instance.
consumption of both systems.

Finally energy sources and energy mix are kept out

Although, you might consider that notes and coins
of scope for a total focus on the most fundamental
are not an electronic form of money and shouldn’t
questions: energy input & energy ef ciency.
be accounted for, yet this form of central bank
money is used in retail electronic payments such as
Our research follows the journey of a classical
automated teller machines, or electronic point of
monetary and payments transaction from money
sales terminals. Therefore they are accounted for in
issuance (yearly renewal) to its transit into an ATM or
electronic payment transactions.
a cash point, followed by its acceptance in payment
as cash. Then we switch to card payments and the
In order to estimate how much energy per year is
energy of schemes like VISA and Mastercard, and
required yearly to print and mint we need to
ePoS acceptance, then we continue the journey
estimated the total number of coins and banknotes in
through the banking system up the stream to central
use worldwide and the current renewal rate. These
bank clearing and settlement mechanisms. Finally we
numbers are not easy to estimate given the lack of
will consider the important update of classical
information from certain countries, the differences in
payments into instant payments. For a Bitcoin
currencies values and consumer preferences for coins
transaction we will consider simply the total
or banknotes which differ largely from one country to
hardware used in mining and processing and
another. To succeed in this critical challenge and to
compute the exact energy consumed by the
reduce the error margin we’ve used 2 different
Blockchain PoW. Later will we also assess the
sources and methodologies to narrow down the
important improvement brought by Bitcoin
evaluation.
Lightning.
According to the ECB there are today 28.67 Billion
At the end we will compute and compare the energy
banknotes and 141.97 Billion coins in circulation in
per a single transaction in both systems: Classical vs
the eurozone representing respectively €1.587 Billion
PoW then Instant vs Lightning. Then we will propose
and €31.426 Billion. After the COVID period the
an energy ef ciency equation that will allow us to
demand for cash increased to reach a total of 16% of
arbitrate on both capacity , speed and energy
the GDP of the eurozone in 2022 counting 342.56
consumption per transaction of both systems.
million persons using the euro. According to Central
Bank of India there were about 124.36 Billion
1 INTRODUCTION 10 ENERGY OF BANKING OFFICES banknotes and 122.99 Billion coins in circulation in
2021 in India. The Federal Reserve Bank published
2 SCIENTIFIC APPROACH 11 ENERGY OF BANKING COMMUTE
that there are 67.68 Billion banknotes in 2022
3 METHODOLOGY 12 ENERGY OF BANKING IT labeled in dollar and worth $2,750.27 Billion. For
coins in the USA, there were about 28 Billion coins
4 SCOPE 13 ENERGY OF INTER-BANKING
in circulation in 2016 but with 15 to 20 billion coins
5 ENERGY OF PRINTING & MINTING 14 RESULTS OF CLASSIC PAYMENTS minted yearly according to the US Department of
Treasury.
6 ENERGY OF ATMS 15 ENERGY OF BITCOIN

7 ENERGY OF CASH IN TRANSIT 16 COMPARING TX LEVEL Based on the above numbers and by extrapolation
according to population we reach a total of 842.6
8 ENERGY OF CASH AT EPOS 17 LIGHTNING VS INSTANT
Billion banknotes in circulation worldwide. To
9 ENERGY OF CARD PAYMENTS 18 CONCLUSION con rm our estimation we’ve checked with a second
source that estimated at 576 Billion banknotes
Content Summary [T1]
worldwide in 2018. Several central banks con rmed

Bitcoin: Cryptopayments Energy Ef ciency 5/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
 a growth rate of 12% to 16% per year of banknotes consumes about 37.2 KWh per page. This allows us
printed since 2018. This leads to a count of notes to state that the printing [1b] can consume at least
between 1015 Billion and 1210 Billion banknotes 8,163 TWh/year. About 83% of this energy is
according to this second source. This con rms our consumed in making the paper that goes through the
range of 842.6 Billion banknotes as a safe lower press. Only 11.1% are needed to print, cut, collate.
bound estimation with an average growth rate of Since we are not taking into account the paper
11.57% per year in the last 4 years. manufacturing but only the running operations of
transforming the paper into banknote this reduces
For coins in circulation applying the same the energy value accordingly.
extrapolation leads us to ~1,507.7 Billion coins
worldwide in 2022. We conclude that globally the Energy(PrintNotes) ≈ 906 TWh/year [1e]
ratio of coins to banknotes is ≈ 1.79× although this
ratio differs largely around the globe between ~1× in
India to ~4.9× in the EU. According to“T he United States Mint’s 2011
Sustainability Report” the US Mint’s total energy
What is important for us is the renewal rate of this consumption in 2011 was 192,906,111 KWh. This
cash mass. It’s estimated that the renewal rate of includes consumption of natural gas, diesel, lique ed
banknotes is ~26.04% per year. This leads to 219.42 petroleum, electricity and steam. In 2011, there were
Billion banknotes per year being printed to replace 8.7 billion coins minted by the US Mint. This gives
the worn notes taken out of circulation and to us an energy cost per coin of ≈23.53 Wh/coin, and
answer new demand. a total energy consumption for yearly minting
worldwide today of about 2.49 TWh/year. This is
Coins have a slower renewal rate. A circulating coin excludes the energy consumption of metal mining
has an estimated 30 years lifespan or more and today and transport and covers only the minting process
the US Mint issued 15 billion coins in 2021 giving a and transformation into central bank coin.
renewal rate of about 11.54% for the dollar and
2.52% per year for coins based on European Central Energy(MintCoins) ≈ 2.5 TWh/year [1f]
Bank’s data. Different countries might have larger
increase rates but we base our calculations on a
renewal rate of ~7% and this leads to a global In conclusion the total energy consumption of
minting rate of 106 Billion coins per year. So the printing and minting cash is
results are as follows:
Energy(Print&Mint) ≈ 908.6 TWh/year [1]
Count(notes) ≈ 842.57 Billion notes [1a]

New(notes) ≈ 219.42 Billion notes/year [1b] Note that on yearly bases, minting is using less energy
than banknote printing because of a very large
Count(coins) ≈ 1,507.7 Billion coins [1c] demand on banknotes compared to coins and
because coins require no maintenance and are not
New(coins) ≈ 106 Billion coins/year [1d]
recycled, their lifecycle virtually ends at production,
while banknotes deteriorate quickly and are
Let’s now estimate the energy consumption of [1b] frequently renewed. We’ve excluded from the
and [1d]. evaluation the energy consumption of the mining of
the metal coins and the manufacturing of paper
Energy consumption of secure paper money printing notes according to our methodology. Finally we’ve
is dif cult to estimate precisely. Paper fabrication, also excluded the initial printing and minting of total
printing, cutting, collating, counting, testing and cash in circulation and only evaluated the new cash
binding is a complex and heavy industrial process (we entering into circulation yearly.
will account for cash in transit later). But an easier
approach that can give us the lower bound, is by
estimating the energy consumption of industrial
magazine manufacturing. We’ve calculated that
printing 35,000 copies of a 96 pages magazine

Bitcoin: Cryptopayments Energy Ef ciency 6/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
 ENERGY OF ATMS ENERGY OF CASH IN TRANSIT

We’ve computed and veri ed in our study that the Now let’s assess the physical cash management
total number of cash dispenser machines worldwide energy consumption and focus on cash-in-transit
is estimated to be 4,823,564 ATMs. This number (CIT), the physical transfer of banknotes, coins,
can be reached based on country-by-country studies credit cards and items of value from one location to
of the number of ATM machines per 100,000 another. The locations include cash centers and bank
individuals. We’ve veri ed this by double-checking branches, ATM points, large retailers and other
central banks sources of 35 representative countries premises holding large amounts of cash, such as
covering more than 5 billion persons on all ticket vending machines and parking meters.
continents and based on of cial central banks Companies such as Loomis and Brinks are the main
reports. Our calculations reached an average of 60.7 providers in this sector and are representative of the
ATMs per 100,000 persons in the total serving the workload necessary.
world’s population of 7,939,000,000 people. This
estimation takes into account the drop of ATMs Loomis has more than 200 cash processing centers
usage in several countries worldwide. equipped with technology to count, authenticate, and
check the quality of banknotes and coins. We will
For a small bank an ATM machine has an average ignore the veri cation and authentication machines
daily power18 of about 250W maximum according to consumption as an additional simpli cation of our
a rst source. We consider 230W per ATM on model and focus only on the transit energy
average for our calculations and evaluate that the consumption. Loomis handles up to 50 million
energy consumption of world ATMs to be around banknotes per day in the processing centers and has
9.72 TWh/yr. A second source is Diebold, the ATM 6000 secure transport vehicles. Brinks has 1100
vendor estimates that an ATM has an average operations facilities, and a eet of 13,300 vehicles.
consumption of 1620 kWh/yr leading to 7.81 TWh/ These vehicles consume signi cant amounts of
yr. We’ll consider an average of 8.77 TWh/yr. energy. A diesel armoured car can be consuming
~3.3 kWh per kilometer. Note that kinetic energy is
But in practice, ATMs are not used alone, they only 30% of the input energy required for the truck,
require two air conditioners and lighting (that can be the remaining is lost mainly in the exhaust. Estimates
in a branch or out of bound of a bank branch). A computed an average of ~35 litres of diesel every
medium air conditioner consumes about 900W. 100 km. One litre of diesel fuel has an energy of ~10
Since running one air conditioner for 24/7 reduces kWh. This con rms the estimate around 3.5 kWh/
its life span, two AC are used working19 alternately to km.
achieve full time coverage. By taking the average
consumption of all ATMs and including only 1 AC CIT is a complex process with multiple steps. Cash
unit this leads to a more realistic range of 46.8 vehicles can be transporting banknotes and coins
TWh/yr for all ATMs worldwide. from cash centres to banks, or from retailers to bank
for instance. Almost all the eet is used daily, CIT
Energy(ATM) ≈ 47 TWh/yr [2a] companies optimise the number of vehicles for these
rotations without counting the energy cost for their
maintenance as an additional simpli cation. In the
This estimation does not take into account: server majority of the time an additional step is added to
side consumption, cash handling, or any the CIT process and a banalised vehicle belonging to
maintenance interventions on ATMs. Therefore it the bank, drives along the armoured car thus almost
can be seen as a minimal energy requirement to run doubling the number of vehicles involved but not to
world ATMs today. all the path of transit.

It is a very dif cult task to estimate the number of

armoured trucks for CIT worldwide. A good
educated guess is to take into account the total

18 These estimations ignores for now the energy consumption on the server-side of ATMs Manager by Diebold Nixdorf or NCR for instance
19 Note that these AC units are not used in the bank branch itself by only dedicated to the ATM

Bitcoin: Cryptopayments Energy Ef ciency 7/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fl
fl
fi
fi
fi
fi
 number of banknotes in circulation worldwide, see payment instruments used at the POS and P2P in
[1a]. We’ve estimated that the velocity of banknotes terms of the volume (count) of transactions to be
is ≈ 1.5 × , meaning that a banknote circulates and about 73% of all payment transactions including
returns back to its starting point (ex. an ATM) 1.5 POS (and 48% of the transactions’ value). We
times per year. compute the average number of POS and P2P
transactions per person per day, per country, to be
Loomis eet in the USA has 3000 vehicles (50% of about 1.14 cash Tx/day/person in Europe with large
their world eet) and process 25 million banknotes disparities between countries cash appetite: Greeks
per day in the US. This leads to an average capacity Italians and Portuguese make 1.6 cash Tx/day/
per CIT vehicle ~3,041,667 banknotes per year. person (compared to 0.5 for Dutch and Estonians).
Taken into consideration the velocity of banknotes Based on the 1.6 gure as an average worldwide we
and the total number of banknotes worldwide we extrapolate to world population, this leads to a
estimate that there are about 6.03 million CIT 3.3 Trillion cash transactions on electronic point of
vehicles in the world including a safe estimate of 1 sale per year. But since Europe has fewer cash
bank car for each 2 armoured trucks counting 5.4 transactions than the rest of the world the real
million trucks. When we take a moderate hypothesis number can be considerably higher.
that an armoured vehicle transports the cash for ~40
km per day over 220 days per year, and the banalised PoS(CashTx) ≈ 3.3 Trillion Tx/yr [3a]
bank car accompanies the truck only for half of the
path our estimations lead to a total cash-in-transit
power of: The total energy for a cash peer-to-peer transactions
in the economy is more dif cult to estimate since they
are private by design, and not all of them are
CIT(Vehicles ) ≈ 6 million vehicles [2b] accounted for electronically. But it’s important to
Energy(CIT) ≈ 166 TWh/yr [2c] account at least for the part that is processed by
electronic point of sales desks because the cash
management has a high cost effort for merchants too,
This gure does not account for processing centres not only banks. Today Visa alone serves 100 million
and employees managing the cash and the merchants worldwide, a number in the rise largely
distribution from central banks. This excludes also driven by government initiatives to promote cashless
the energy consumption for the maintenance payments. An educated guess of how many
activities for these vehicles, so [E] should be merchants accept cash payments only, can be the
considered as a minimum energy requirement for majority of the very small shops around the world
global CIT activity. that still do not accept card payments. Based on
ECB estimates that 27% of vicinity payments are
card transactions and 73% are cash at ePOS we can
ENERGY OF CASH AT EPOS
estimate that 370 million merchants accept cash
using a PoS electronic system worldwide. If we
Physical cash payments are private by nature and consider that a PoS cash desk works 8 hours per day
(without electronic traceability) are harder to count. on average and uses as little energy as a PoS terminal
In our research we’ve found that the share of cash about 111.6 W (see references thereafter) this leads to at
transactions at an electronic POS is more frequent least 72.75 TWh/yr for electronic cash desks at PoS
than the cash distribution by banks and even more working 220 days per year 8 hours per day and
frequent than electronic vicinity payments at shops. counting only 1 e-cash ePoS per merchant.21
In Europe, cash transactions are on average about
68.1% of all vicinity payment transactions at POS20.
Energy(CashPayments) ≈ 72.8 TWh/yr [3b]
Globally, les notes and coins represent more than
50% of transactions in most OECD countries. To Energy(ePosCashTx) ≈ 22.03 Wh/Tx [3c]
verify this initial estimate we refer to the European
Central Bank report that estimates the share of

20Going from 88% in Malta to 77% in Germany to as low as 34% in Netherlands (59% in France)
21We will not account for cash transport by merchants between branches and by client since this could impact highly the results without
being completely in scope for comparing with Bitcoin payments.

Bitcoin: Cryptopayments Energy Ef ciency 8/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fl
fl
fi
fi
fi
 ENERGY OF CARD PAYMENTS Energy(SchemeTx) ≈ 11.49 Wh/Tx [4a]

After estimating cardless PoS electronic cash

To complete the estimation, a card payment
management, we will estimate in this paragraph the
generally originates at a point of sale (POS) terminal
energy consumption of card payment transactions.
in the case of vicinity payments24. We’ve estimated
The card payment leader Visa has 4 data centers
the global installed base of POS terminals to be
located in Central US, East US, UK, and Singapore,
about 207 million units in 2020. Then we’ve
with a private communication network of 10 million
computed that the minimum POS energy
route miles (400 times earth circle22). Although Visa
consumption per transaction ≈ 0.9401 Wh/Tx. In a
keeps con dential the exact energy required by its
different and more accurate method, the average
datacenters, it is possible to compute that Visa’s they
POS terminal energy consumption25 is ≈ 111.66 W.
require on average 305 MW per data center to
If a terminal works 8 hours/day on average and is
operate [4c]. These calculations are based on Visa’s
online 80% of working hours, we compute that the
annual report (Visa Green Bond Report — July 2021)
total POS terminals energy consumption is more
stating that Visa’s datacenter energy consumption
precisely 54 TWh/yr for the 207 million terminals.
stayed stable between 2017 and 2020 totalling 446
million kWh. This estimation leads to an energy
consumption by Visa ≈ of 2.7 TWh/yr [4a]. Visa’s Energy(POS) ≈ 54 TWh/yr [4b]
market share can be estimated to be about 15% of
total cards in the world. This can be obtained Based on [4] and [4b] we conclude that card
through Visa’s declaration that is processes 3.8 Billion payment transactions consumes 71.71 TWh/yr.
cards and we know that the total number of payment
cards was 25.2 Billion cards in 2021. Now we can
Energy(CardSchemes) ≈ 71.71 TWh/yr [5]
extrapolate [4a] and see that the total card schemes
payments datacenter consumption is ≈ 17.72 TWh/ Energy(CardSchemeTx) ≈ 46.51 Wh/Tx [5a]
yr to operate all card payments worldwide.

Note that the difference between electronic cash at

Power(Schemes) ≈ 17.72 TWh/yr [4]
PoS and ePoS is that one is only at cash desks energy
consumption [3b] and the ePoS is for energy
So based on Visa numbers we can extrapolate that payment terminals accepting card payments [4b].
the total payment cards generate 1.54 Trillion And these are 2 different hardware serving different
transactions per year (Tx/yr) [4d] or about 48,891 means of payments at the point of sale.
Tx/s moving USD 86.2 billion/yr. This leads to an
average of 56 USD/Tx (in card payments) [4e].
ENERGY OF BANKING OFFICES
As a reminder these transactions are not nal, they
are most of the time online authorisation requests to We need to complete the estimation with the end-to-
the Issuer on behalf of the Acquirer23 and the end work ow through banks and other cross nancial
payment will complete later with the collection, institution actors such as clearinghouses and Swift
clearing and settlement transactions with central like service providers. This does not include all the
bank money. We will estimate the energy of these services provided by banks such as insurance, loans
steps later in our study. As a result, card payment or trading, but focuses on the accounts and payments
transactions (which is only an intermediary step of management services. The banking industry counts
the end-to-end payment transaction) require globally more than 25,000 banks around the world [6a].
about 11.49 Wh/Tx in energy to process through a Banks manage globally a high number of branches
card scheme like Visa or Mastercard for instance. that also consume a large amount of energy to

22 That is 5 096 800 Km of cables only between servers at the datacenter

23 In card payments Issuer is the issuing bank of the card and the Acquirer is the bank of the merchant acquiring the payment on his behalf
24 Since online payments originate in the same way as Bitcoin cryptopayments and e-commerce card payments they are kept out of the
scope of the calculations.
25 Source: Energy Losses Due to Imperfect Payment Infrastructure and Payment Instruments by Oleksandr Melnychenko, Oct 9, 2021

Bitcoin: Cryptopayments Energy Ef ciency 9/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fl
fi
fi
fi
fi
 operate. Commercial bank branches according to total of 46,250,597 employees working in the
International Monetary Fund, is about 14.145 banking sector worldwide [7a].
branch per 100K individuals. Based on world
population we can estimate that there are 1,122,972 Given a list of about 100 top payment service
bank branches [6b]. providers worldwide, the average number of
employees per PSP is about 12000. We have
estimate the total number of employees working in
Count(Banks) ≈ 25,000 [6a]
payments industry to be more than 1.2 million
Count(BankBranches) ≈ 1.123 Million [6b] em p loyees wo rld w i d e ran gi n g f ro m ATM
manufacturers, ePOS manufacturer to payment
switches and processors etc.
Energy consumption of banking branches can be
evaluated through the average number of kilowatt
hours per square meter for a commercial building. Bankers ≈ 46.25 Million employees [7a]
According to the Department of Energy (DOE), this
Payments ≈ 1.2 Million employees [7b]
electrical energy is approximately 242.2 kWh/m2.

Traditionally, bank branches have ranged26 in size These estimations account for direct employees and
from 371.6m2 to 557.4m2. We have veri ed from an ignore the somewhat signi cant numbers of external
different source, a major French bank, that an providers such as security and guards, CIT
average size bank can consume about 21% of its employees, consultants etc.
energy on of ces and branches27 and according to
banks roadmaps of improving its energy ef ciency of It is important to distinguish only the count of
its of ces and branches, it is targeting an average of employees in [7a] working at least partly on
137.14 KWh/m2 in 2024. This con rms the range payments and accounts management. We consider
and although banks worldwide are improving their that employees working on security, fraud,
carbon impact using their resources large disparities maintenance, risks, marketing, accounting,
worldwide still remain between them. We will c o m p l i a n c e, a n d o t h e r c ro s s s t r e a m s o r
consider the average of 242.2 kWh/m2 in 2022 administrative functions fall in the same domain of
worldwide. This leads to an energy consumption by payments and cash management since they are
total bank of ces and branches worldwide of 151 essential for these services. Although branch
TWh/yr, excluding for now IT services and banks employees work also on loans and insurance sales but
data centres. they are still essential in the cash management,
distribution and management of credit cards, in
addition to the execution of certain wire transfers
Energy(BankBranches) ≈ 242.2 KWh/m2 [6c]
and book keeping.
Energy(BankBranches) ≈ 151.6 TWh/yr [6]
It’s legitimate to reduce [7a] by subtracting banking
functions solely related to loans, insurance and
trading. Worldwide there are about 9.6 million
ENERGY OF BANKING COMMUTE
traders, but not all of them work in banks. For
example at the main large banks in France the
The number of employees working in banking can Société Générale has 351 employees out of 133,000
be estimated by extrapolation from representative employees worldwide, and 353 employees at the
countries where such statistics are available of cially BNPP are traders out of the 193,319 total employees
by central banks. Based on major American bank count. On average we can see that 0.22% of the total
reports, central banks in China, India, Eurozone, and headcount is in trading which is negligible 103,365
Japan we can estimate that 1,850 employees work per traders working in banks. In another approach we
bank on average around the world. This leads to a can consider that about 25% of banking
headquarters work on loans, insurance and trading,

26 We’ve based the calculation at 6000 square feet instead of 4000 to account for HQ and additional dependencies other than the bank
branch of ces for public.
27 Source: Internal of cial information of one of 12 major French bank according to their program for improving energy ef ciency.

Bitcoin: Cryptopayments Energy Ef ciency 10/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
 but almost 100% of branch employees fall into cash such, NAS storage etc. we will consider the lower
and payments operations at least part time. bound of28:
Therefore this demonstrates that the strictly non-
banking and non-payments related employees can be Count(BankServers) ≈ 300 servers/bank [8a]
considered to be at the margin and that [7b] can be
reduced by a factor of 0.54% to focus only on
employees concerned in the fundamental banking In terms of annual energy usage, a two-socket server
and payment services. As a safety measure we will may use approximately 1,314 kWh/yr (which is
ignore [7b] and reduce slightly [7a] to reach a lower simply just powering it on) to about 2,600 kWh/yr
bound of banks employees. [8e]. IBM servers’ idle energy usage is related to the
number of Central Processing Unit (CPU) sockets
and has remained static since 2007 at 365W for two
Count(BankEmployees) ≈ 46 Million [7c]
socket servers (Shehabi et al, 2016; Shehabi et al,
2018). But in reality a server in a data center facility
The average commute distance in transport per requires more power. A more accurate estimate
employee is known to be approximately 24.14 km might come from calculating how many servers could
one-way. This is an average distance between large be used with a given energy capacity. If a similar
city centres, suburbs and far country sites. According high tier data center has an 850 MW capacity, and
to major french bank internal numbers, commute each rack was using 25 kW of power, that institution
energy is 57% due to car travels, 26% air travel, 17% could operate 1,768,000 servers. This leads to a
train totalling 4400 Km/year per employee on minimum required energy of 0,481 kW/server that
average or 20 km/day con rming our estimate then consumes 4,212 kWh/yr [8b]
above.
To count the total number of servers used in banking
An employee's transport energy consumption is 7 we proceed as follows in a different approach to
kWh per km per employee. Based on [7c] This leads validate our estimations. While Amazon and
to 3,420 TWh/yr for all employees’ daily round trips Alphabet devote 12% and 20% of their operating
considering 220 work days per year on average and costs respectively to IT, banks are devoting 29% of
ignoring air travel as a simpli cation of commute their operating costs to IT spending. This informs us
especially that banks policies are restricting nowadays that banks’ IT budgets are 2,42 × times higher than
travel by airplane to a minimum. GAFA and other industries [8c]. There are 100
million servers that are currently being used all
Energy(Commute) ≈ 3420 TWh/yr [7] around the world. A substantial number of these
servers are owned by Google and Microsoft. In total,
we’ve estimated that bank data centres use 7,500,000
servers [8d], 7.5% of all servers worldwide which is a
logical minimum gure based on [8c].

Let’s adopt a different approach. We have veri ed

ENERGY OF BANKING IT with a major French banks that it possess 2
datacenters consuming 1250 KW each. Leading to
Banks also possess data centres either in the vicinity 21.9 GWh/year. Using [6a] this extrapolates to 547.5
internally or outsourced or on the cloud. The most TWh/year for all banks datacenters. But not all
precise way to calculate the energy consumption of a banks are major banks. For instance in France the
bank’s data centre is to estimate the average number ratio is 3.13% major banks and this gives us about
of servers used per bank. Based on our knowledge 783 major banks worldwide out of the 25000 banks.
and our consultation with small and large banks we This allows us to estimate that major banks consume
can estimate that each bank (small banks) has a about 17.15 TWh/year on their datacenters. While
minimum of 300 servers varying between large IBM smaller banks consumes much less but are much
mainframes to rewalls, database servers, routers, more in numbers. Based on [8d] and [8b] we
core banking and online banking, backup servers compute that the minimum bank data centres energy

28 This is the average for small banks, so in reality the numbers are at much higher

Bitcoin: Cryptopayments Energy Ef ciency 11/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
consumption is ≈ 31.6 TWh/yr with 85 MW per transactions like Faster Payments in UK and SEPA
datacenter on average for all banks worldwide. Major SCTInst Scheme in Eurozone. We can either use a
banks consumes about 55% of this energy spent. methodology of estimating energy consumption of
encryption and transfer of large data transfers
Energy(BankDatacenters) ≈ 31.6 TWh/yr [8] between banks and CSM or use a simpli ed
approach using individual level transactions. Based
on [10c] and both [4a] and [10a], we can evaluate
the lower and upper bounds of energy consumption
ENERGY OF INTER-BANKING
for CSM messages to be between 36.2 TWh/yr and
79 TWh/yr. It’s safe to consider the average30 equal
Inter-banking communications using nancial to 57.4 TWh/yr. The rationale of this approximation
messages for wire transfers clearing and settlement is that certainly a large clearing batch le consumes
are also required to complete the study. Swift hugely more than a single card authorisation call as
operates three data centers - one in Zoeterwoude, found for a single Visa transaction [4a], Power(CSM)
Netherlands; another in Culpeper, Virginia, United > 36.2 TWh/yr. And it’s since the transactions for
States; and a third in Thurgau, Switzerland. It also CSM are grouped by thousands, to tens of thousands
has a Command and Control center in Hong Kong. each transaction will consume much less than a 1MB
Not all banks are connected to Swift but only 11,000 email as in [10a] so the Power(CSM) < 79 TWh/yr.
banks. We can conclude from the above sources that So the average is clearly near the real value.
Swift represents 44% of this use, based on [6a].
Power(CSM) ≈ 57.4 TWh/yr [10]
Using [4c] we estimate a minimum of 21.29 TWh/
yr of the energy consumption of Swift-like messages
in data centres alone. Finally, banking employees use personal computers,
as well as banking software relying on backend
Energy(FinDataCenters) ≈ 21.3 TWh/yr [9] servers usually on the cloud. On average, there are
20 deployed computers per server leading to
2 312 530 servers in the backend for banking as well
Clearing and settlement mechanisms (CSM) are used as 46 million personal computers for bank employees
by banks to complete payment transactions like EBA (See [7c])
in Europe and STET in France for instance. As a
reminder sending a simple email on the internet Using [8c] and [11a] this translates into 6.01 TWh/
requires 25 Wh of energy [10a]. As a sample, STET, yr for total backend servers and cloud energy used by
the French CSM processed in 2020: 16.74 billion banks for AWS, Azure and other SAP like SaaS. In
transactions per year [10b]. The total card volumes29 addition, desktop computer consumes about 600
according to Banque de France is 49% of total kWh/yr [11c] leading to 27.60 TWh/yr.
payment transactions. Based on [4d] this means that
the total number of world payments is 3.146 trillion
Count(BankITServers) ≈ 2.3 million [11a]
Tx/yr, including card and non-card payments, such
as wire transfers and direct debits, requiring clearing Count(BankPC) ≈ 46 million [11b]
in most cases.
Energy(FrontToBack) ≈ 33.58 TWh/yr [11]
Count(PaymtTx) ≈ 3.146 Trillion Tx/yr [10c]

CSM messages are cleared in batch le mode in

general, yet now they are becoming instant payment

29 Number of transactions not amounts

30 As a reminder a single payment clearing transaction requires several API calls between small bank to the primary bank to CSM then to
the central bank and back to the bank and the payer and the payees accounts, batch le calls also uses streaming of les over the internet
which consumes much more energy than a simple and small single transaction authorisation the quantity of data in EMV norm of card
payments contains much less data then a batch grouping tens of thousands of transactions per hours or day. So it’s a moderate estimation to
only use [10a] and [4a] instead of n times for n API calls energy costs. Here we used an approximation because such energy consumption
differs very highly between banks.

Bitcoin: Cryptopayments Energy Ef ciency 12/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
 RESULTS FOR CLASSICAL PAYMENTS given that not all hardware is produced in equal quantities and
readily available”. So there’s plenty of room for
improvement on their work and that’s what we will
Finally, we have completed the evaluation of the
undertake in this paper with a completely different
energy consumption for all the classical monetary
methodology.
and payment systems. In conclusion, we can estimate
that the total energy consumption is the sum of
The best and most scienti cally precise method is to
intermediary results in TWh/yr.
count Bitcoin miner nodes and hardware units and
then based on the required computing power (PoW
Energy(ClassicSystem) ≈ 4981 TWh/yr [12] dif culty) of the installed base of miner units, we can
precisely estimate the kWh actually used by each
Given the large differences in scale between [1][6][7] mining unit available on the Blockchain. Today
and the other energy cost centres, it is important to 100% of mining units are a special hardware model
clarify the legitimacy of including them. As a called ASIC (Application-Speci c Integrated
reminder, old telephone center buildings and their Circuit). Therefore, we can exclude today Bitcoin
telephone booths around the world got transformed mining through CPU or GPU since they are out of
through telecom industrialisation and were the network or extremely marginal.32
eventually replaced by electronic telecom switches for
ef ciency, scalability and better services. The same After China’s mining ban, Bitcoin Mining Map
way can be applied to the services domains that shows that USA became recently the leading Bitcoin
Bitcoin aims to cover. Note also that [1][6][7] can be mining country with 35.4% of global hash rate
considered a lower bound since we didn’t take into power of the Blockchain. It’s important to escape
account any central banks nor all the additional listing all currently in use ASIC hardware according
registered electronic money issuers and payment to their pro tability. This approach requires to cater
service providers data centres (such as Stripe, or for electricity prices and we do not need to use this
PayPal for instance). inaccurate path in our research. A better path is
simply hardware ef ciency: that is Watt consumed
Next we will evaluate Bitcoin PoW energy. per Terahash and the release dates of each model.
Non pro table miner units are most probably today
switched off, because by de nition the miner will be
ENERGY OF BITCOIN literally loosing money if they switch them on,
depending on electricity costs. That said, we’ve
Let’s now analyse the Bitcoin blockchain PoW energy veri ed with industrial miners that they are still using
consumption, excluding its layer 2 for Bitcoin certain older ASIC models considered today non
Lightning for now. The most referenced assessment pro table based on average electricity price. That’s
work is Cambridge Bitcoin Electricity Consumption possible because certain industrial miners negotiated
Index (CBECI). According to Cambridge, Bitcoin very low cost of energy rendering pro table older
power is supposed to be equal to 144.82 TWh/yr models.
[13] with a lower bound of 53.29 TWh/yr and a
higher bound of 356.83 TWh/yr. This large range Let’s call i the index of a mining unit model mi out of
seems more as a guess31 work by Cambridge than a Nt models at time t. We call the Power Ef ciency πi of
precise evaluation and these numbers are a mining unit m i , the amount of power required to
continuously used to criticise the PoW of Bitcoin. reach a rate of 1 tera-hash.
Cambridge acknowledges using different hypothesis
of electricity prices for pro tability estimations and Po w e r (m i ) [14]
πi = (in W per TH)
“uses simplistic weighting of pro table hardware” yet Ha sh r a t e (m i )
Cambridge is aware that “assuming that all pro table
equipment is equally distributed among miners is unrealistic

31 Cambridge Index speaks of constructing a “best-guess”

32CPU and GPU mining represent considerably less than 0.000000001% of mining power and they are completely inef cient and not used
actively in mining. We will only consider CPU and GPU mining in our research to compile initial mining data in the rst 5 to 6 years of
PoW mining

Bitcoin: Cryptopayments Energy Ef ciency 13/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
 Let’s start by building the mathematical model and count Ci (t ) of installed units for each model over
theory of Bitcoin PoW mining. Let at any point in time. But since sales data are not available and the
time t: Blockchain does not register the mining model this
approach seems impossible. Yet we’ve found that the
Mt = {m 0 , m1, . . . , mNt} |M | = Nt miner models increase of the total hash rate ΔHt of Bitcoin can
give us a precise indicator of the growth of the hash
Where M is the set of miner unit model m i and Nt is power of the total installed miners: At any point t in
the total count of models available at any point of Nt

time t since the mining started in 2009 until today in time, H (t ) = ∑ hiCi (t ) . [15] where H (t ) is the hash
i=0
2022.
rate available on Bitcoin at time t. This data can be
reliably read from Bitcoin Blockchain in near real
Let Rt = {r0, r1, . . . , rn} |R| = |M |
time. The hashing power is estimated from the
Such that ∀m ∈ M, f : M → R give us the release number of blocks being mined in the last 24h and
date and allows us to determine the age αi of the the current block dif culty. More speci cally, given
mining unit in order to determine later its market the average time T between mined blocks and a
share of the hash power and its energy consumption. dif culty D on Bitcoin PoW, the estimated hash rate
Give a time t, we can model all miners attributes per second H is given by the for mula
D
using this approach: H (t ) = 232 × . Where T ≈ 10 min but in the
T
calculation of hash rate the real value on the
Mmodel = {m 0 , m1, . . . , mNt}
blockchain are used and that can be for example less
Mrelease = {r0 , r1, . . . , rNt}
than 10 min. Therefore the hash rate although
Mhash = {h 0 , h1, . . . , hNt} calculated, is extremely precise for our work up to 10
Mi n e r s (t ) →
Mpower = {p0 , p1, . . . , pNt} minutes periods, while we are working on monthly
Mef f iciency = {π 0 , π1, . . . , πNt} rates and numbers over 13.3 years period.
Mcount = {C0 (t ), C1(t ), . . . , CNt (t )}
The hashing Dif culty is a measure of how dif cult it
is to mine a Bitcoin block, or in more technical
Fo r e a c h m i n e r w e c o n s i d e r t h e s e t terms, to nd a hash below a given target. A high
m i → {ri , αi , h i , pi , πi , Ci (t )} where Ci (t ) is the count dif culty means that it will take more computing
of the model m i at a given time t, online and power to mine the same number of blocks, making
contributing to the Bitcoin Blockchain PoW. ASIC the network more secure against attacks. The
miners life expectancy is between 3 to 5 years and dif culty adjustment is directly related to the total
we’ve veri ed this information with the mining mining power in the Total Hash Rate (TH/s). When
industry directly. So in our research we’ve accounted the Bitcoin hash rate increases or decreases by ΔH (t )
for Nmax = 91 miner models with a release date this is due to the fact that the installed hardware park
between July 2014 and May 2022 (αma x ≤ 5 years increased in the total count of mining units across all
period). Other older models can be safely considered models m i . This delta increase or decrease will be
out of the Bitcoin network today (or not impacting distributed among older and newer hardware
the results). released.

Nma x = 91 m o d el s These considerations lead to one equation with up to

ri ∈ [rmin, rma x ] = [2014.07, 2022.03] 91 or more variables making it impossible to
αi = t − ri , αma x ≈ 60 m o n t h s compute each Ci (t ) . Yet the total hash rate is
Mi n e r s → h i → T H /S distributed over the available models Nt. This fact can
pi → Wa t t lead us to a rst average calculation but we can
πi → Wa t t /T H achieve a much better evaluation. We note that for
Cm (rmin ) = Cm (rma x ) ≈ 0 , ∀i ∈ [1, 91] t ≤ Ju l y 2014 all old miners are not online anymore,
i i
therefore it’s allowed to estimate them as one uni ed
virtual model. Let’s call m 0 the virtual miner
The most precise way to evaluate the Bitcoin PoW hardware starting since the r0 = rmin = 2009.02
energy consumption is to calculate the most accurate (month 1 of Bitcoin going online). Since this virtual ASIC
miner can be considered alone in the market until

Bitcoin: Cryptopayments Energy Ef ciency 14/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
 r1 = 2014.07 (month 68), it’s now easy to compute the H (61) 15,943 T H /s
C0 (61) = = = 88,570 miner units33
count of m 0 for all the period C0 (t ) ∀t ∈ [1, 68] using: h0 0,18 T H /s

H (t ) Once a new miner model m1 is released to the

C0 (t ) = ∀t ∈ [1, 68] total months [16]
h0 market at least 1 to 3 months are required for it to be
part of the Blockchain hashing power, and will follow
In the example above we see that in this model the a growth rate replacing gradually m 0 market share
virtual m 0 with its 0.18 TH/s is allowed to have a and hash rate until it leaves the installed park of
count of units less than one (until Feb 2011) since it miners and goes of ine ~60 months (5 years) later.
simulates all older models: 2 ASIC models, some Using its life time, it will share the hash power with
GPU and CPU models. And since m 0 and all these newer models m i launched after its market entry.
models combined are no longer online, it’s still a very
accurate start to build on for the next released Since the sales are not known and the average
models. distribution is not precise enough, we’ve
demonstrated that Bitcoin hash rate variation ΔHt is
So m 0 equations become as follows: directly proportional to the sum of all ΔCm (t ) that
corresponds to the total miners installed and taking
Nma x = 1 m o d el into account new hardware, minus older miners gone
r0 = rmin ∈ [2009.02, 2019.05] of ine. This leads to ΔHt = Ht − Ht−1, ∀t ∈ [1, 160]
αma x ≈ 120 m o n t h s months being directly proportional to: Δ Count(miners
h 0 = 0.18T H /s at t):
m0 → p0 = 360W
π 0 = 2000W /T H Nt Nt Nt

∑ ∑ ∑
H(t)
Ci (t ) − Ci (t − 1) = ΔCi (t ) ⇒
C0 (t ) = i=0 i=0 i=0
h0
∀t ∈ [1, 160] m o n t h s Nt

∑ i
ΔHt = h × ΔCi (t )
i=0
Note that exceptionally the maximum age of m 0 is
10 years instead of 5 years. It is legitimate to allow And the total count of mining units being at a given
for new models to replace it in the market and Nt
because the last miner issued more than 5 years ago time = ∑ Ci (t ) miners
i=0
is the Bitmain AntMiner S1 with 0.18TH/s model
and released in Nov, 2013, this model is used as m 0
Since hi and pi are given by the manufacturers for all
unifying all older ASIC, GPU and CPU inef cient
models, we need to nd the most accurate approach
hardware. At the same time, Bitcoin total hash rate
to calculate ΔCm (t ) for each one of all Nt miner
was millions of times smaller growing from H(1)=
models at a given34 time t.
0.0000044 TH/s in 2009.02 to H(58) = 2,559 TH/s
in 2013.11, an increase of 581 million folds in about
If the distribution of new hardware to the market is
4 years. This con rms our approach for m 0 and its
made evenly (which is not) and before we apply
0.18TH/s capacity as the uni ed model of all older
additional improvements to this method, it’s possible
hardware no longer online since 2019.05.
to solve an approximation of the many variable
equations using the series based on m 0 information :
For example, the virtual number of miner units is:
ΔHi ΔHi
H (20) 0.00417 T H /s Ci (t ) = Ci−1(t ) + giving us ΔCi (t ) = [17]
C0 (20) = = = 0.0232 miner units Nt h i Nt h i
h0 0,18 T H /s

At this stage this is a virtual model count. Miners here are not nodes. Nodes are often pools of mining hardware of ASIC models. At
33

month 61 we can estimate that 88,570 miner units were online actively participating in PoW on Bitcoin nonstop.
34 Overclocking miners will increase the hash rate and increase the consumption. We did not take this into account in the model since the
Power Ef ciency remains relatively the same thus not affecting the total energy consumption of Bitcoin in total but only requiring marginal
less hardware consuming marginally a little more power but ending with an equivalent energy consumption globally.

Bitcoin: Cryptopayments Energy Ef ciency 15/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fl
fi
fi
fl
fi
fi
fi
fi
 Let’s now nd a very accurate solution for the t − μi 2
2 ( σi )
equation. For m 0 , the C0 (t ) is determined 100% 1 −1
γi (t ) = e [18]
accurately thanks to blockchain data until the release σi 2π
date of m1 at r1 = 68 . Until α 0 = 68 , m 0 is the only
mining virtual model in the market. This will allow
Where π = 3.14159265359... and e an exponentiel
us to compute a rst (but unsatisfying) estimation of
function related to Euler’s number e = 2.71828…
all mining units over time. In order to improve this
then σ is the standard deviation and μ is the mean
model to the maximum extent, we need to nd a
value of the distribution (median). Since we know
precise distribution model of miner units installation
that a miner age αma x ≈ 60 months, we can consider
usage and gone of ine.
that μi ≈ 30 months on average for any miner for
Bitcoin.
In economy and in physics new products entering a
market follow globally the normal35 distribution law
We still need to determine the σi . The standard
in general. We will not undergo a full demonstration
deviation is a measure of the dispersion of a set of
here, but let’s state that this has been veri ed for
miner model m i . A low standard deviation indicates
many products in logistics and here’s a quick logical
that the values tend to be close to the mean μi = 30
veri cation for the miners products:
months, while a high standard deviation indicates that
the values are spread out over a wider range. Since
Once a new mining unit is released, it takes a
we are looking for an αma x ≈ 60 months as the wide
certain delay to be pre-sold, sold, manufactured,
spread of the normal curve, this solves to a
stored, shipped and delivered, thus initiating the
σi ≈ 6.5 m o n t h s ∀i > 0. The interpretation of this
start of the bell shape with an exponential growth.
value is that between −σi and +σi , a period of 13
Mining market information shows a delay of 1 to 3
months, 68.2% of the total model m i sales are in
months to start a growth penetration phase of the
production and online on the blockchain. And
product.
between −2σi and +2σi (26 months period) 95.4%
are in production. This leaves a long tail outside the
Then once it becomes largely available in the
range ± 2σi (more than 2 standard deviations) for
market, it grows relatively exponentially in sales
entering the market and exiting the market on the
and installation before its growth rate decelerates
borders of the curve, and this was veri ed with
and then gets limited due to 2 factors: market
mining centres still using almost 5 years old
saturation, price to power ef ciency (πi), and new
hardware. Given in our model σi = 6.5 months and
better models entering the competition limiting the
μi = 30 months we can now compute the percentage
demand on the previous models.
each product sales and installation. We need now to
Once the interaction of several competing models correlate the product penetration rate with different
occur, the previous model decelerates its growth to competing models and determine the most precise
halt at a maximum marketshare then the curve count of each model in time, that is to solve
starts reversing while the new models grow Ci (t ) , ∀i ∈ [1, Nt ], ∀t ∈ [1,160] months.
exponentially and the older model starts getting Nt
out of market following the reverse process is Since the ΔHt = ∑ hi × ΔCi (t ) and since we know
sensibly symmetric, until the same process occurs i=0
in loops also for the newer models. the value of C0 (t ) ∀t ∈ [1,68] , it’s possible to
compute one by one the series Ci (t ) based on Ci−1(t )
The maximum age on average of any m i is 5 years of using the proportion of each model sales given by
full-time run thus giving us a knowledge of the Gauss γi (t ). We’ve demonstrated that the solution for this
curve span. This important piece of information is complex equation is in [19] below.
crucial to solve the complex equations. A normal
distribution curve requires 2 variables to be modelled H (t ) = C0 (t )h 0 + . . . + Ci (t )h i + . . . + CN (t )hN ⇒
the:
Nt

∑ i
H (t ) = C (t )h i
i=0

35 Also called Laplace-Gauss law, Gauss Law or simply bell shape curve

Bitcoin: Cryptopayments Energy Ef ciency 16/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fl
fi
fi
fi
fi
fi
 H (t ) γi (t )
Verifying Ci (t ) = × ∀i ∈ [0,Nt ] ⇒ H (160) ≈ 204 × 1018 H /s [21]
hi N t γ (t )
∑ j=0 j
To t a lCo u n tM (160) = 3,990,685 miners [22]
Nt Nt
H (t ) × γi (t ) EBitcoin (160) ≈ 88.95T W h /y r ± ϵ [23]
∑ h × ∑ Nt γ (t ) ∑ j
H (t ) = × h i , since γ (t ) = 1
i=0 i j=0 j j=0
Cumulative miner models count = 92 models [24]
Nt Current miner models online today = 80 models [25]
∑
⇒ H (t ) = H (t ) × γi (t ) ,
i=0 πM = 49.7W /T H on average of current miners [26]

Nt
Which is true since ∑ γi (t ) = 1. And that demonstrates that To see the evolution of PoW Power Ef ciency over
i=0 time see πPoW graph in the diagram [D5] above.

H (t ) × γi (t ) The importance of the count of hardware per model

Ci (t ) = ∀i ∈ [0,Nt ] ⇒ ∎
hi is that it allowed us to calculate that there are ≈ 4
million miner units of the different models today
worldwide grouped in the 15,636 reachable Bitcoin
t − μi 2
2 ( σi )
H(t) −1 nodes. Note that our approach do not require any
Ci (t ) = e
h i σi 2π mapping per country nor uses electricity pricing
mi → [19] directly or indirectly in anyway.
∀i ∈ [0, Nt ], ∀t ∈ [1, 160]
σi = 6.5 m o n t h s
μi = 30 m o n t h s Since the power of a hardware is determined by the
manufacturer and veri ed by the mining community
we can consider its value precise with a very low
This important mi equation allows for an accurate error margin (less than 2%). The error margin of
description of Bitcoin energy consumption based on bitcoin energy consumption can originate from 2
the precise number of miner units for all miner sources:
models. The equation results are con rmed for all
calculations up to 0.00000001 precision at any point Our Ci (t ) estimation and the precision of the pi in
in time t during the 160 months period. Watts provided by the manufacturer37. In our
approach we do not need to consider that the
Bitcoin Blockchain Explorer indicates that the hardware up time rate per year (~ 99.9%) since
current miners network hash rate ≈ 204 million tera- we’ve computed the exact real ASIC miners
hashes per second36. And we can read on the uptime consumption.
blockchain all H(t) values daily since the genesis
block. Varying the values of σi and μi generate results
deviating by ϵ < 5 % comforting the precision of
Now we can nally compute the energy consumption our calculations. In contrast, the error margins of
of Bitcoin PoW using : Cambridge index are very large with
Nt ϵCBECI ∈ [−63.2 % ,146.4%].
∑ i i
P (t ) = p C (t ) [20]
i=0 This updates our [23] to:

Where P(t) is the electrical power of Bitcoin’s PoW in

EBitcoin (t o d a y) = 88.95T W h /y r (ϵ < 5%) [23]
Watt at time t. We were able to compute the number
of each installed miner units during 160 months for
each mi model. Here are the values today at t = 160 Note also that not all the 92 ASIC miner models
months. used are solely for Bitcoin mining many of them are

36 That is 204,000,000,000,000,000,000 cryptographic hashes/second

37Note that overclocking ASIC miners may improve the hash rate capability and increase the required energy consumption value of the
miner unit but the power ef ciency remains relatively constant thus not affecting the nal result especially that all us bound by the Bitcoin
total hash rate. So this can produce a variation less then the error margins of our Ci(t)

Bitcoin: Cryptopayments Energy Ef ciency 17/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
 Bitcoin Hash rate in TH/s, months [0, 160]
0 00
00
50
22

0 00
50
87
16

0
00
00
25
11

00
00
25
56

1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126 131 136 141 146 151 156

[D1] Bitcoin hash rate in according to the Blockchain data

0
00
00
35

0
00
25
26

0
00
50
17
00
50
87
0

1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126 131 136 141 146 151 156

[D2] Count Ci (t ) of each one of the 92 ASIC miner models contributing to PoW over [0, 160] months
0
00
00
90

0
00
50
67

0
00
00
45

0
00
50
22

1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126 131 136 141 146 151 156

[D3] Total count of mining units over time (accumulation of the count for the 92 ASIC miner models) over 160 months

PoW Miners Power Ef ciency in W/TH

00
22
50
16
00
11
0
55
0

1 12 23 34 45 56 67 78 89 100 111 122 133 144 155

[D4] Calculation of Ci (t ): Count of miners for 92 ASIC models [D5] Evolution of PoW total miners Power
(left to right columns) in [0, 160] months (top down rows) Ef ciency πi of all models overs 160 months
See [26]

Bitcoin: Cryptopayments Energy Ef ciency 18/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
 not that ef cient for Bitcoin and are mostly used to COMPARING AT TRANSACTION LEVEL
mine other cryptocurrencies. Based on this principle
alone we can see that the [23] is actually an upper
We can conclude that the cryptopayment system of
bound. And if we take this argument into account,
Bitcoin PoW consumes at least ~56 times less energy
miners with higher πi will be more used in mining
than the classical electronic monetary and payment
Bitcoin and this will improve the actual Bitcoin
system.
energy consumption further.

Hypothetically if all mining units were replaced by Eclassic ≈ 56 × EBitcoin [27]

the most ef cient ASIC (having the lowest πi possible
today), Bitcoin energy consumption would drop to Let’s compare now the work and power of the 2
πi = 29.5W /T H instead of [26]. This leads to the systems. It’s not enough to compare only the total
same H(t) work with a large drop of Bitcoin energy energy usage of both system. We have to compare
consumption by 40% to only 52.8TWh/yr. So it’s now the energy ef ciency per transaction level and the
quantity of work and power involved in these 2
systems.
Annualised Bitcoin PoW Energy Consumption in TWh
0
14

Let’s start by comparing the energy consumption at a

0

single transaction level. For Bitcoin, the current block

12

size is between 1 MB to 1.52 MB and hosts today

0
10

about 2,591 Tx per block. A Bitcoin block now have

80

a theoretical maximum size of 4 MB and a more

60

realistic maximum size of 2 MB39. The exact size

depends on the types of transactions. So the
40

maximum capacity of bulk processing per block can

20

be about ~10,380 Tx/block. This result is obtained

0

1 11 21 31 41 51 61 71 81 91 101 111 121 131 141 151 using a variable size of a bitcoin transaction between
303 and 454 KB/Tx (from median to average).
[D7] Energy consumption of Bitcoin PoW We’ve computed that Bitcoin can process up to
between [0, 160] months 544,879,300 Tx/yr and currently is processing about

possible to run all Bitcoin network with 52.8TWh/yr Co u n tcurrent (B i t c o i nT x) ≈ 136.22 × 106T x /y r [29]
today without triggering the hash rate adjustment the
hash rate required for PoW. Note that this is not an Co u n tma x (B i t c o i nT x) ≈ 544.88 × 106T x /y r [30]
actual lower bound but minimum required energy
Pe a kreal (B i t c o i n Ca p a c i t y) ≈ 4.32T x /s [31]
today and the hardware uses actually 88.95 TWh/yr
to do the same work38. Pe a k ma x (B i t c o i n Ca p a c i t y) ≈ 17.3T x /s [32]

E n e r g yavg (1T x) ≈ 653kW h /T x [33]

Minimum possible energy consumption today
[23a]
EMin (B i t c o i n) = 52.8T W h /y r E n e r g ymin (1T x) ≈ 115kW h /T x [34]

EPoW (160) ≈ 1,180,220 kWh/block (today) [33a]

PPoW (160) ≈ 10.154G W [33b]

136.22 million Tx/year (operating at 24% of its
capacity). So on average a single Bitcoin Tx requires

38Note also that commute energy consumption of employees running the ~ 15,636 nodes is negligible compared to 46 million banks
employees without counting the very high number payment service providers employees. In addition most of the nodes manager work is
done remotely in monitoring and remote maintenance. We didn’t include physical maintenance of banking and Bitcoin since it’s in the
disadvantage of the classical payments system.
39This difference is due to the fact that Bitcoin uses now a maximum “weight limit” of a block. Block weight is a measure of the size of a
block, measured in weight units. The Bitcoin protocol limits blocks to 4 million weight units, restricting the number of transactions a miner
can include in a block. Four million weight units is equivalent to 4MB of data, meaning the maximum size for a block is now 4MB.

Bitcoin: Cryptopayments Energy Ef ciency 19/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
 today 653 kWh/Tx but can be executed in 115kWh/ comparing apples to oranges. A bitcoin transaction
yr with more adoption of Bitcoin in cryptopayments. and a classical system transaction do not have the
same number of steps, pre-requisites and most
Note that [33a] is obtained through the current total importantly do not get settled at the same duration.
power of PoW 10.154 GW in [33b] and a currently A Bitcoin transaction gets settled in near real time
low Median Block Con rmation Time = 6.974 within 10 minutes on average40. While in a classical
minutes/block instead of 10 minutes/block. payments transaction the settlement occurs in about
1 to 5 business days i.e. up to 7 days41. Most of local
transactions are usually settled in T+2 which is in 2
0

days. Cross border payments settle slower due to

80

kWh/Tx
additional barriers. Classic payments can be
6
68

settled up 1008 times slower than a Bitcoin

PoW energy per Tx between 460 to 653 kWh/Tx transaction. This is the case for cross-border
1
57

payments for example. They are ≈ 2% of total

7
45

payments worldwide but increasing extremely fast

due to massive adoption of online payments and
3
34

mobile payments with global marketplaces and

9

immigration working force. But in the larger case, we

22

can consider that the settlement duration of classical

4
11

payment transaction is on average 2 days. A classic

0

payment transaction is on average 288 times

00

00

00

00

00

00

00

00

00

00

00

00

0
10
17

24

31

38

45

52

59

66

73

80

87

94

slower than a Bitcoin transaction. [35]

10

[D7a] Current energy consumption of a single Bitcoin In physics power is the energy consumed in a given
transaction using PoW in kWh per Tx versus number of duration, and the work is a form of change in energy
Tx per block in [1700, 10100] Tx/block that also when divided by time gives power. So let’s
compare work done by Bitcoin and the work done by
The minimum power per Bitcoin transaction can the classical electronic payment and banking system.
occur when the maximum capacity of a block is Based on [0]
used. Note that [33] and [34] cannot be extrapolated
to larger volumes since Bitcoin transactions are The classical electronic system consumes at least 1.58
grouped in blocks and a single block can contain up kWh/Tx on average (see [12] and [10c]) but
to ~10K Tx. We will see below the method that completes the work in ~48 hours on average. Bitcoin
Bitcoin can scale above this block limit using Bitcoin consumes between 115kWh/Tx up to 653 kWh/Tx
layer 2 called the Lightning Network. currently and nishes the transactions in ~10
minutes on average (currently in 418.44 seconds).
The current monetary payment system is at least
5,775 times bigger than Bitcoin in terms of payment It’s also important to understand the meaning of the
transaction volumes [30] and [10c], and had 60 years average 1.58 kWh/Tx for the classical system. This
more time to get optimised and to scale yet consumes average is based on all 3.146 Trillion Tx/yr that the
~56 times more energy than Bitcoin PoW does. banking and payments system processes while
consuming 4981 TWh/yr (see [10c] and [12]). And is
Comparing energy ef ciency per transaction of the computed as the lower bound estimation after several
two competing systems is not a direct calculation of simpli cations. It’s clear that the banking system as a
the total transactions count per year over the total whole can sometimes consume much more or much
energy consumption. Although this might seem less energy per transaction depending on the
logical to compare energy on a single transaction, different nature of the payment act: cross border,
doing so is incomplete because it would be card, or non-card, instant payment, cash transaction,

40The current Median Con rmation Time on Bitcoin Blockchain has recently dropped to ~7 minutes. Bitcoin PoW mining dif culty is
adjusted every 2016 blocks (every 2 weeks approximately) so that the average time between each block remains 10 minutes.
41Swift GPI hasn’t been adopted yet but will allow in the future for faster cross border transactions within minutes too. Central Bank Digital
currency experimentation (Donbar Project by BIS) tested a different approach using multi-CBDC between central banks directly. But these
systems are not yet in production. See also Swift How long do wire transfers take?

Bitcoin: Cryptopayments Energy Ef ciency 20/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
 η (function of Bitcoin PoW energy ef ciency per transaction)
6,0
vicinity or online transactions. Similarly it’s
important to understand that the only sound 5,3
estimation is the energy consumption of Bitcoin is
4,5
per block. Since a block can contain 0 to 10,380 Tx
yet it consumes the same amount of energy per block 3,8
(see [33a]). Therefore the minium Bitcoin transaction 3,0
in PoW is currently Pmin ≈ 115kW h /T x and the
monthly average is Pavg ≈ 653kW h /T x. See [34] 2,3

1,5
First conclusion, Bitcoin is not used to its full η ∈ [1.0, 1.4]
0,8
potential. Since we can increase the
transaction volumes ×4 without increasing its 0,0

1 400
2 000
2 573
3 200
3 800
4 400
5 000
5 600
6 200
6 800
7 400
8 000
8 600
9 200
9 800
energy consumption. Block size is underused
today and Bitcoin adoption can grow without
increase in energy. When this maximum limit is Transactions per block
reached this will be the highest volume capacity
[D8] PoW η energy ef ciency vs Classical system,
Bitcoin can handle, and that’s why we cannot
starting from the current lowest count of Tx/block up
extrapolate energy growth to be converted to more
to the maximum block capacity (with current miner
throughput above this cap. Bitcoin solved this limit
models mix)
by introducing Lightning Network at a layer 2 that
we will cover in the next paragraphs.
At a single transaction level, Bitcoin today is [36]
Based on [0], we’ve de ned that: the same work done η ≈ 1.2 × more energ y ef cient than the
by both systems is moving money through a Classical System with a range of ηB ∈ [1.0, 1.4]
displacement in time, instead of moving a physical
ηma x = 5.7 today [36a]
object through space. The proper methodology in
order to compare apples to apples is to compare the
energy consumption relative to the settlement time of “Eta" η can be analysed as energy ef ciency or action
the 2 systems. Let’s call Compared Energy ef ciency as per [35]. That is temporal ef ciency
Ef ciency or energy conversion ef ciency (η) a number multiplied by work ef ciency or the ratio of action A
without a unit obtained as a ratio of the useful work per transaction of both systems.
output of an energy conversion system compared to
another system, here Bitcoin compared to the global At its current capacity, Bitcoin PoW can use 412
monetary and payment system. times more energy per transaction than the electronic
system and can nish the same work 413 times faster
Pc tc2 ΔWc . d tc A with a median block con rmation time of 418.44
η = = = ηe × ηd = c
PB tB2 ΔWB . d tB AB seconds today (~7 minutes/block). At a block rate of
10 minutes, Bitcoin PoW is at least 288 times faster.
ΔWc dt
Where, ηe = and ηd = c . We can conclude that PoW layer 1 of Bitcoin is
ΔWB d tB
today ηBitcoin ≈ 1.2 times more energy ef cient
Note that dt is the displacement in time as in [0], e is than the electronic system at the transaction
for energy, c for classic system and B for Bitcoin. ηe, is level. Yet this ef ciency is not yet used to its full
the energy ef ciency per transaction and ηd , is the potential today. If bitcoin adoption doubles then η =
duration ef ciency per transaction (distance in time). 2 becoming twice more energy ef cient than classical
Note also that A = P . t 2 is called action in physics. system at a single transaction level. Today at current
The action is the momentum of the transaction block size and if the blocks are lled to their
times the displacement it moves through time and maximum capacity ηmax = 5.7× better energy
has dimensions of energy × time, and its unit can be ef ciency than the classical system. In the future, old
in joule-second (like the Planck constant h). miner models will get eventually replaced by newer
more ef cient models and without increasing the
Ac hash rate nor the transaction count per block, ηBitcoin
Energy Ef ciency is: η = ηe × ηd = [35]
AB ≈ 6.71 times more energy ef cient per transaction

Bitcoin: Cryptopayments Energy Ef ciency 21/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
 based only on PoW. Then 52.8 TWh/yr would hop hop hop hop hop hop
A B C D E F G
consumed by PoW instead of 88.95 TWh/yr today, 1 2 3 4 5 6

see [23a].
Today Lightning is in production and live but is still
in its early adoption with 36,852 nodes (14,950 nodes
Energy Capacity Tx Time Ef ciency with public IPs) and 83,601 open channels. Lightning
Energy
network capacity is today 141 millions $ with an
TWh/yr Tx/yr kWh/Tx Min η average node capacity of 8,183$ (0.213 BTC). An
instant payment transaction on Bitcoin costs only 1
Classic 4981 3.146 1.58 2880 1× satoshi as a median base fee and takes a fraction of a
Trillion second to nalise.
Bitcoin 89 133 653 ~7 1.2×
Current Million
A Bitcoin Lightning node can be modelled using a
raspberry pi with an SSD usually used by the nodes.
Bitcoin 89 544.88 115 ~7 5.7× Such a unit consumes about 5W if both CPUs are
PoW Million busy which is not the case all the time. A transaction
is processed in less than a second roughly in ≈500
[T2] Bitcoin Energy Ef ciency compared to the classical milliseconds per transactions as the duration of actual
System. (Bitcoin current average and Bitcoin current maximum processing. Given the estimation of 6 hops:
capacity PoW)
EL2 (L i gh t n i n g) ≈ 6 × 5W × 0.5s = 0.00416667W h
per transaction42 or 7.5W of power, if the
BITCOIN LIGHTNING VS INSTANT PAYMENTS transactions are treated in a single mode. This is
about 480,000 times less energy than a classical
Let’s take into account now Bitcoin Lightning payments transaction.
network compared to Instant Payment (IP) schemes
that are drastically increasing the nality time of the EL2 ( Lightning ) ≈ 0.000004167 kWh/Tx [35]
transaction respectively of both systems. The
Lightning Network has an important capability to
Let’s compare this result to an instant payment
scale up exponentially the transactions throughput
transaction for instance in the eurozone (SEPA
above Bitcoin layer 1, yet it does so without growing
SCTInst scheme). A payment transaction is initiated
in a proportional rate to the energy input.
at an online banking account relying on several
It’s important to understand rst how Lightning banking servers doing compliance, online banking
transactions work. Lightning leverages existing backend, core banking layer 1 and 2, instant
Bitcoin transaction channels between peer to peer payments servers such as Payment as a Service PaaS
payers and payees to group additional Bitcoin and the server for instant payments that calls the
Lightning transactions in a single Bitcoin PoW CSM to reach the central bank instant payment
transaction on the main Blockchain. For example, if system to check 1st the reachability of the payee’s
Alice A wants to pay Georges G, 1000 satoshi bank, thus making several calls among at least 5 to 10
(=0.00000001 bitcoin ≈ 4$ today) Lightning will nd hops just for the reachability checks. Then ounce the
the fastest open channel that’s already executing reachability is OK between payer and payees banks,
transactions on its path to include the amount and the instant payment order is given and the processing
make the payment instantly on this channel. From A goes through the same hops in addition to more
to G the transaction can be direct between 2 nodes if depth reaching the core bank main layer (we do not
they have an open prepaid channel between them, if account for the noti cations of the payer and the
not and the transaction needs to go to the other end payee of the execution of the transaction since these
of the globe, it’s assumed that ≤ 6 hops are required are sensibly the same also for Lightning). The total
from A to reach G: number of hops can be about more than 20 hops
between heavy duty servers in data centres and
mainframes in addition to centralised systems at the

42This is a minimal energy cost today without counting for smartcontracts added value services above Bitcoin Lightning that will become
possible thanks to new protocols under development such as TARO and Watchtowers.

Bitcoin: Cryptopayments Energy Ef ciency 22/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
fi
fi
 central bank linking everyone. For eurozone instant Trillion Tx/yr that might seem to correspond to
payments SCTInst scheme, the clearing and 99,759 Tx/s. But for a large part these transactions
settlement mechanism CSM can be through an are bulk payments and not individual peak capacity.
intermediary like STET in France, linking banks to Note from [4d] that the maximum capacity today of
the TIPS system at the European Central Bank. Note card payments authorisation is limited to ≈48,891
also that an instant payment transaction is not always Tx/s. At the same energy input in both systems,
a nal transaction, it guarantees the nality in compared to Bitcoin, the Lightning Network can
advance since it uses mirror accounts at CSM of the handle 1,000,000 Tx/s in a single channel that is
banks accounts at central banks. The complete 20.45 times more capacity than the classical system
nality is usually delayed especially when the central and still 345,600 times faster or at least 14 times
bank system is not available. faster than Instant Payments43.

Note that comparing Bitcoin Lightning to Faster Lightning is 14 × faster than Instant Payment [37]
Payments or Instant Payment Schemes without
counting the underlying channel closer on Bitcoin
using PoW is a valid approach. The rational is that a And total Energy consumption of Bitcoin and
Bitcoin Lightning transaction is nal at layer 2 and its Bitcoin Lightning is sensibly the same:
an option to write it later on the layer 1 of Bitcoin
with PoW or to close the channel. Also saying that an EL1L2 (B i t c o i n) = EL1 + EL2 = EL1 + ϵL2 ≈ 88.95T W h /y r
instant payment in classical system can take 7
seconds instead of 48 hours is sometimes wrong Theoretically if we consider that Instant Payment is
because in reality it can takes in certain cases 3 hours fully adopted worldwide even for cross border
to 24 hours to settle completely with central bank payments, the maximum scale up capacity is
money while on Bitcoin Lightning it is nal after a estimated by Swift to be 1000 Tx/s. That is a
fraction of a second and on the worst case it can take maximum throughput of 31.53 Billion Tx/year. This
10 minutes (optionally) with a channel closing on the shows a scalability ceiling for current electronic
main Blockchain. centralised payment systems limiting the adoption of
instant payments for about only 1% of current global
Based on the preceding analysis of Lightning and payment transactions. In order to increase this limit
Instant Payments we can consider that instant important hardware and architecture changes are
payments energy consumption is equivalent to the required from core banking, online banking, payment
same old classical system energy consumption since it hubs, Swift GPI to central bank systems.
uses the same hardware and banking infrastructure
but only prioritises and accelerates certain In order to evaluate the precise total average energy
transactions part of the work ow. In addition the ef ciency of an Instant Payment transaction, we can
means of payment by itself as instant payment is still omit all cash, CIT and ATMs related energy and
not yet widely adopted in payments globally since it keep only banking, PSP, and Interbanking energy
is still missing additional complementary services consumption. This leads to 3,860 TWh/yr for a
such as Request to Pay, link to card payment maximum capacity of 31.53 Billion Tx/year. Note
initiation and PoS acceptance solutions. Finally an that this lowered total energy consumption means
instant payment is today a local payment and is not that there are no more banknote and coins, which is
available for cross border transactions. On the other not realistic since all central banks are promising the
hand Bitcoin Lighting is still also in its beginning of opposite. Since IP cannot replace all electronic
development with certain issues like stuck payment transactions we are forced to maintain the
transactions that cancels after a longer time than 1 cash servicing energy. We will consider the additional
second. As we see both innovations are competing energy required to service IP negligible.
but still not 100% stable and not deployed to a large
scale. Energymin ( IP ) ≈ 3860 TWh/yr [38a]

Comparing payments throughput, we saw in [10c] Energyavg ( IP ) ≈ 4981 TWh/yr [38]

that the classical system has a capacity of 3.146

43 Which is technically sometimes not nal in 7 seconds and require clearing.

Bitcoin: Cryptopayments Energy Ef ciency 23/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fl
fi
fi
fi
 Energytx ( IP ) ≈ 158 kWh/Tx [39] owe you” (IOU). His bank owes to him this amount of
money at the central bank. This is an important
Capacity(IP) ≈ 31.53 Billion Tx/yr [40] difference between Bitcoin and the classical money
system. A Bitcoin transaction between a payer and a
payee is a direct transfer of the cryptoasset: bitcoin
Note that [38] is only theoretical since today cross
functioning as cryptocurrency without the need of
border transactions are not instant payment ready
any trusted third party. The nature of the
and the total number of instant payment Tx is below
cryptocurrency Bitcoin is cryptographically different.
maximum capacity of [40]. Average energy of a
It is a token or real value and not money as debt or a
single IP transaction [39] is between 120 and 158
promise, on the contrary it is nal with a direct
kWh/Tx based on [38a] and [38] respectively and
ownership of the intrinsic value of the asset a feature
the theoretical today’s maximum in [40].
that the electronic money system does not offer
consistently except in banknotes and coins. That’s
By applying the same method to a Lightning
why comparing Bitcoin to electronic money and
transaction, we will not consider the single layer 2
payment system is not comparing 100% similar
energy and its 0.000004167 kWh/Tx. For Bitcoin we
systems. Yet in this paper we’ve endeavoured to
will also take both layers leading to:
compare their common promise only from an energy
ef ciency point of view, ignoring all additional
EnergyL1L2 ( Lightning) ≈ 88.95 TWh/yr [41] features of both. For instance Bitcoin is also a
Energytx ( Lightning) ≈ 0.00282059 kWh/Tx [42] programmable form of money with less complexity
of its value chain and participants, allowing the
Capacity(IP) ≈ 31.54 Trillion Tx/yr [43] commoditisation of the whole classical payment
system and allowing additional services such as smart
contract and programmability. In contrast, Instant
CONCLUSION Payments is not, and require additional services such
as Request to Pay scheme and also a link to a card
scheme which are not built into the classical system
Today when we transfer 1 dollar from a payer to a
yet.
payee, there is no direct real44 transfer of value done
between the two. This is due to the fragmented
Globally, results prove that Bitcoin uses 56× less
nature of electronic monetary system today. What
energy than the classical system even without
should take place during the transaction in the
the inclusion of Lightning and Instant Payments to
electronic system is a change of ownership of the
the benchmark and without comparing to any other
asset called money, but in reality it is a conversion of
cryptopayments consensus mechanism using proof
commercial bank money supervised by the central
of stake for example.
bank. The payment executes a burn of this private
electronic money conserved by the bank of the payer Today at a single transaction, level Bitcoin Proof of
and then a different operation of earn is performed Work is in average 1.2× more energy ef cient than a
for the equivalent amount with a different privately classic electronic payment transaction and can go up
issued electronic money by bank of the payee. The to 5× with more adoption or natural replacement of
settlement of this burn and earn is achieved through a older mining units with more ef cient hardware
distinct transaction45 executed between the bank of already available.
the payer and the bank of the payee using their own
accounts at the central bank. When Bitcoin Lightning and Instant Payments are
included in the benchmark, and by simulating that
The payee, although he received the payment, will they are used to their highest capacity in both
never have a direct claim on it. It’s an amount that systems we nd that a Bitcoin Lightning transaction
he’s is lending to his bank, equivalent to a promise: “I is in average 345,000 times faster than classical

44 Money in cash form (banknotes and coins) are indeed directly exchanged between Alice and Georges but they are not 100% direct
exchange of money. They are a direct exchange of central promise of the face value on the paper or coin guaranteed by the central bank. So
there’s no intrinsic value being instantly exchanged.
45 This transaction groups all the Alice to Georges transaction but bundled with all transaction between these 2 banks after the netting of

their values

Bitcoin: Cryptopayments Energy Ef ciency 24/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fi
 system and 14 times faster than an instant payment operate trillions of transaction per year in relative
transaction. In addition Bitcoin Lightning scales slow speed between 2 to 7 days to complete. This
far higher than Instant Payments with a over optimisation and specialisation causes it to be
theoretic capacity of 31 Trillion Tx per year for fragmented, fragile and less capable of scaling up
Lightning compared to 31 Billion Tx per year for today to instant payments. While Bitcoin has a
Instant Payments. This capacity limitation is mostly higher power output ratio and is capable to scale very
due to the Swift like cross border systems limitation ef ciently using Lightning Network, thanks to the
in throughput at ~1000 transactions per second. In PoW layer of its main Blockchain.
our estimation we’ve used 1000 Tx/s as a global
maximum for instant payments, but in reality
different regional or country systems may be unable
to reach that capacity for instance, European Central
Bank instant payment scheme (TIPS) is estimated to
process up to a maximum average of 500
transactions per second. As a consequence, if both
systems are used to their maximum capacity, the
energy cost of a single instant payment transaction
becomes much higher than a classical payment and
requires ~158 kWh/Tx. Bitcoin scales better and
manages to decrease drastically down to 0.00282
kWh/Tx on his 2 layers combined (so including
PoW).

In conclusion, Lightning at a single transaction level

allows Bitcoin to become 194 Million × more
energy ef cient than a classical payment and
up to 1 million × more energy ef cient than
an instant payment Tx.

We can observe that the classical system is over

optimised to consume less energy per transaction to

Total Total Comparing Tx Comparing

Energy Capacity Tx Energy Tx Duration Energy Speed Energy Ef ciency

TWh/yr Tx/yr kWh/Tx Seconds ηE ηS ηEE

Classical
4 981 3.14 trillion 1.58 172 800 1× 1× 1×
System †

Bitcoin PoW 460 290.6 ×

88.95 133 Million 418.44 413 × 1.2 ×
Current to 653 to 412.5 ×

Bitcoin PoW
52.8 545 Million 97 600 61.2 × 288 × 4.71 ×
Best Possible

Instant
4 981 31.53 Billion 158 ~7 99.8× 24 686 × 247 ×
Payment ‡

Bitcoin L1+L2 † 194 Million ×

88.95 31.54 Trillion 0,00282 ~0.5 0.0010576 × 345 600 ×
Lightning ‡ 1 Million ×

[T3] Figures of Instant Payment & Lightning: both systems used to their maximum capacity to demonstrate full potentiel and
scalability. Hypothetically every Tx is switched to its fastest mode in each system

Bitcoin: Cryptopayments Energy Ef ciency 25/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
 REFERENCES “Global POS Terminals and Wireless M2M Market Report
2021: POS Terminal App Stores are a Potential Game-
Changer” — Research and Markets, December 2021
“2020 Annual Review and Consolidated Financial Statements”
— SWIFT Annual Review “Glossaire Des Cryptopaiements” — France Payments Forum, 2020

“A Map of Forex Traders: Where Are the 9.6 Million Online “Guidelines for the next generation of Real-Time Retail
Traders?” — brokernotes.co, 2022 Payments Systems (RT-RPS)” — Swift, 2015

“Armored Vehicles Are Key to the Cash-in-Transit Industry” — “How long do wire transfers take?” — Swift
The Armoured Group, 2022
“How Many Dollar Bills Are in Circulation?” — TitleMax, 2017
“Assessing IBM's POWER8, Part 2: Server Applications on
OpenPOWER” — Johan De Gelas, Sept 2016 “How Many Watts Does An Air Conditioner Use?” — Bardi, 2021

“Average number of individuals employed per bank in Europe “How much is the total power supply consumption of an ATM
(EU28) in 2018, by country” — Statista machine?” — Quora, 2019

“Banknotes and coins circulation” — ECB, 2022 “How Much Money Is In The World Right Now?” — Go
Banking Rates, Feb 2022.
“Benchmarking Commercial Building Energy Use Per Square
Foot” — IotaComm, August 20, 2020 “How much power does a computer use? And how much CO2
does that represent?” — energuide.be
“Bitcoin, totem & tabou — Que pr sage l’essor des
cryptomonnaies ?” Institut Sapiens, Fev 2018 “Identifying IT Markets and Market Size by Number of
Servers” — Applied Computer Research, Inc., 2011
“Bitnodes estimates the relative size of the Bitcoin peer-to-peer
network by nding all of its reachable nodes.” “Le grand paradoxe – ou pourquoi le règne du cash est loin de
s’achever” — Terra Nova, 2021
“Cambridge Bitcoin Electricity Consumption Index” —
Cambridge (CBECI), 2022 “Leading banks in the United States as of December 31, 2021,
by number of employees” — Statista
“Circulation velocity and how you can in uence it” — De La
Rue, 2018 “Lightning Network Explorer” — explorer.acinq.co, 2022

“Climat: des emails à l'e-book, l'impact du numérique n'a rien “Loomis Our business” — web.archive.org, 2014
de virtuel” — Science & Avenir, Mars 2016
“Loomis vs Brinks Comparison” — macroaxis.com, 2022
“Community Banks Get Creative With Unused Branch Space”,
Independentbanker.Org, January 2, 2018 “Miners pro tability” — asicminervalue.com

“Comparing Bitcoin & Lightning energy usage to the real “Money Minting - The process of producing coins”, CFI, 2022
world” — Co-Authored by Oliver Barratt & Danny Scott, 2021
“Number of credit, debit and prepaid cards in circulation
“Con rmed Transactions Per Day” — blockchain.info worldwide from 2017 to 2021, with forecasts for 2023, 2025,
and 2026” — Statista
“CZ Binance” — twitter.com, March 14, 2022
“Number of individuals employed by credit institutions in
“Diebold Innovation Leads to World's Greenest, Most Power- Europe (EU28) from 2009 to 2019” — Statista
Ef cient ATM” — prnewswire.com, March 2014
“Real-Time Lightning Network Statistics” — 1ML.com,
“Electric trucks like the Tesla Semi are 'pointless both
economically and ecologically,' according to a vehicle-tech “Répartition du nombre de salariés du groupe bancaire BNP
expert” — Johannes Kaufmann and Qayyah Moynihan, Business Paribas dans le monde de 2017 à 2020” — Statista
Insider Deutschland , Apr 3, 2019
“Scaling Renewable Energy” — equinix.com, 2022
“En moyenne, en France, un trader gagne 1 million d'euros par
an” — lepoint.fr, 2014 “Share of cash transactions at POS (points of sale) in the euro
area in 2016 and 2019” — Statista
“Energy hogs: Servers vs. desktops vs. set-top boxes” — Patrick
Thibodeau, JUL 6, 2011 “Société Générale Fact Sheet”, Société Générale, 2022

“Energy Losses Due to Imperfect Payment Infrastructure and “Spatial proximity and distance travelled: Commuting versus
Payment Instruments” — Oleksandr Melnychenko, Oct 9, 2021 non-commuting trips in Flanders” — Kobe Boussauw, Frank Witlox

“Everything you need to know about cross-border payments” — “STET Fact Sheet” — STET, 2021
emerchantpay, 2021
“Study on the payment attitudes of consumers in the euro area
“Fourteen Facts about the US Penny” — Design Life-Cycle, 2014 (SPACE)” — ECB, Dec 2020

“From Home to Work, the Average Commute is 26.4 Minutes” “TARGET Instant Payment Settlement User Requirements” —
— US Bureau of Transportation, OCT 2003 ECB, 2017

“Gestion des stocks et loi normale” — JY Baudot “The carbon footprint of bitcoin” — De Nederlandsche Bank,
Juan Pablo Trespalacios and Justin Dijk, 2021

Bitcoin: Cryptopayments Energy Ef ciency 26/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fi
fi
fi
fi
fi
fl
é
 “The Energy Consumption of Blockchain Technology: Beyond
Myth” — Johannes Sedlmeir, Hans Ulrich Buhl, Gilbert Fridgen, Robert
Keller, 2020

“The Hidden Expense of Energy — Print Is Costly, Online Isn't

Free” — The Scholarly Kitchen, 2022

“The lifespan of money”, Giesecke+Devrient GmbH, 2022

“The State Of Lightning” — Arcane Research, Oct 2021

“Total Hash Rate (TH/s)” — blockchain.info

“United States Data Center Energy Usage Report” — Shehabi

et al, 2016; Shehabi et al, 2018

“Visa Data Centers: The engine rooms of Visa’s global

commerce network”, Visa, 2017

“Visa Fact Sheet” — Visa, 2022

“Visa Green Bond Report” — July 2021

“What is Energy and How Much do You Use?” — Sustainability

Exchange, 2022

“What Is The Bitcoin Block Size Limit?” — Bitcoin Magazine,

Aug 17, 2020

“What is the diesel consumption per kilometer of trucks?” —

Bridgestone Web eet Solutions, Feb 2020

“Which consumes more power: Banks or Bitcoins?” — Zodhya,

April 2018

ABOUT THE AUTHOR

(*) Michel KHAZZAKA (@kneisseh), founder of

Valuechain a Cryptopayments consulting enterprise
and YouTube channel. Computer engineer
specialised for 17 years+ in payments security and
innovation. Lead of Cryptopayments workgroups at
several french payments and inter-banking entities.
An international speaker and trainer, has given
lectures and courses in 5 continents on blockchain
and cryptocurrencies, some of which at Sorbonne
and Paris Dauphine universities in Paris and the
American University of Beirut.

Bitcoin: Cryptopayments Energy Ef ciency 27/27 Michel KHAZZAKA — Valuechain

Electronic copy available at: https://ssrn.com/abstract=4125499

fl
fi