The Cambridge Bitcoin Electricity Consumption Index (CBECI) provides an up-to-date estimate of the Bitcoin network’s daily electricity load. The underlying techno-economic model is based on a bottom-up approach initially developed by Marc Bevand  in 2017 that uses the profitability threshold of different types of mining equipment as the starting point.

Given that the exact electricity consumption cannot be determined, the CBECI provides a hypothetical range consisting of a hypothetical lower bound  (floor) and a hypothetical  upper bound  (ceiling) estimate. Within the boundaries of this range, a  best-guess  estimate is calculated to provide a more realistic figure that approximates Bitcoin’s real electricity consumption.

The lower bound estimate corresponds to the theoretical minimum total electricity expenditure based on the best-case assumption that all miners always use the most energy-efficient equipment available on the market. The upper bound estimate specifies the theoretical maximum total electricity expenditure based on the worst-case assumption that all miners always use the least energy-efficient hardware available on the market, as long as running the equipment is still profitable in electricity terms. The best-guess estimate is based on the more realistic assumption that miners use a basket of profitable hardware rather than a single model.


The CBECI landing page displays two numbers for each type of estimate. 

The first number refers to the total  electrical power  consumed by the Bitcoin network and is expressed in gigawatts (GW). This figure is updated once a day and corresponds to the rate at which Bitcoin currently uses electricity. In other words, it describes the current electricity demand of Bitcoin miners (electricity load).

The second number refers to the total yearly electricity consumption of the Bitcoin network and is expressed in terawatt-hours (TWh). It is an annualised measure that assumes continuous power usage at the aforementioned rate over the period of one year. We apply a 7-day moving average to the resulting data point in order to make the output value less dependent of short-term hashrate fluctuations, and thus more suitable for comparisons with alternative uses of electricity.

Model parameters 

The model considers the parameters outlined in Table 1 below. The following sections will specify how each estimate is calculated and what assumptions have been used. 

Table 1: CBECI model parameters





Network hashrate, mean daily 

The average rate at which miners are solving hash puzzles on a given day 

Exahashes per second (Eh/s) 


Bitcoin issuance value, daily 

The aggregated value of all bitcoins newly issued on a given day 



Miner fees, daily 

The aggregated value of all transaction fees paid to miners on a given day 




Difficulty, mean daily 

The average difficulty level of the hash puzzle on a given day 


Bitcoin market price, close daily 

The fixed closing price of bitcoin as of 00:00 UTC on a given day 



Mining equipment efficiency 

Measures the energy efficiency of a given mining hardware type 

Joules per Gigahash (J/Gh) 

Static: hardware specs from 100+ equipment types, taken from various sources 

Electricity cost 

Global average price of electricity incurred by miners 

USD per kilowatt-hour ($/kWh) 

Static: estimate (assumption) 

Power usage effectiveness (PUE)  

Measures how efficiently energy is used in a data centre  


Static: estimates (assumptions) 


Selecting mining equipment 

In the first years of Bitcoin, mining was mainly performed using general-purpose graphics processing units (GPUs) and  field-programmable gate arrays (FPGAs). This changed considerably when in 2012 the first  application-specific integrated circuits (ASICs) started to emerge. ASICs are specialised hardware specifically optimised for Bitcoin mining that are orders of magnitude more efficient than previous devices used for mining. As a result, it did not take long for ASICs to dominate and eventually displace GPU and FPGA mining.

We have compiled a list of more than 100 different Bitcoin ASIC models designed for SHA-256 operations that have been brought to market since 2013. The list is based on a combination of public resources that list various types of mining equipment and their specifications.1 For the pre-ASIC period from 2009 – 2013, we have chosen hardware based on research conducted by Taylor (2016) and our own observations. For the lack of available performance data in 2009, we select the Intel Core i5-650 model despite it only being released in January 2010.

Versions prior to v1.2.0 were based on the full list of hardware specified above. This approach may have periodically overstated Bitcoin’s total power demand for a variety of reasons. With v1.2.0, the following changes were introduced to address this issue:

  • Consolidation of hardware manufacturers to exclude ‘exotic’ devices by only considering ASIC hardware from the three major manufacturers Bitmain, MicroBT and Canaan, which are estimated to have a combined market share of at least 85%.2 This change removes ‘exotic’ devices with little to no sales from the hardware sample. It is important to note that this does not apply to hardware in our list with a release date prior to July 2014.

  • Introducing a maximum economic lifetime of five years to reflect the finite usability of hardware. The lack of such a constraint may risk overstating the power demand by including nominally-profitable but long-disposed hardware models that are not in use anymore.

  • Remove Coinmetrics’s estimate for Bitmain S7 and S9 hardware share by retroactively returning to an equally-weighted basket of profitable hardware models. Potential issues related to the underlying nonce analysis may have overstated the share of older, less efficient Bitmain S7 and S9 ASIC models in the current hardware mix distribution.

Mining efficiency of each machine type is expressed in Joules per Gigahash (J/Gh): given that real power usage can vary significantly depending on several parameters (e.g. usage conditions, overclocking), the manufacturer specifications have been refined with the help of experts to reflect actual power usage more accurately. The full list is available at and is open to comments and suggestions. Figure 1 shows the evolution of Bitcoin mining equipment efficiency since late 2014.

Note: a 1000W mining device that generates 10,000 Gigahashes per second (Gh/s) has an efficiency of 0.1 Joules per Gh (J/Gh). This chart is based on a list of 100+ SHA-256 mining equipment available at

The profitability threshold 

The key idea underlying the CBECI model is that miners will run the equipment as long as it remains economically profitable in electricity terms. To determine the time periods during which a given hardware type is profitable, we model the economic lifetime of each machine by taking into account total miner revenues, total network hashrate, hardware energy efficiency, and the average electricity price per kWh that miners have to pay.

This results in the following mathematical inequality:

It is worth noting that profitability in this context exclusively considers electricity costs incurred for running the machines: it does not take into account capital expenditures (e.g. acquisition and amortisation costs) nor other operational expenditures (e.g. cooling, maintenance, and labour costs).

The profitability threshold (θ) is then calculated as follows: 

Assumption 1: the global average electricity price is constant over time and corresponds to 0.05 USD/kWh.

Electricity prices available to miners vary significantly from one region to another for a variety of reasons. We assume that on average, miners face a constant electricity price of 5 USD cents per kilowatt-hour (0.05 USD/kWh). This default value is based on in-depth conversations with miners worldwide and is consistent with estimates used in previous research.3 

Note: The CBECI landing page allows visitors to choose different values for the electricity cost parameter in order to explore how electricity prices influence hardware selection and total electricity consumption. 

Assuming a fixed electricity price of 0.05 USD/kWh, we can model the evolution of the profitability threshold over time (Figure 2). While mining equipment with an energy efficiency below 2 J/Gh remained profitable in early 2015, the threshold has substantially decreased over time because of the introduction of newer ASIC generations and a continuous increase in hashrate. Large price spikes occasionally lead to a sharp increase of the profitability threshold (e.g. crypto bull runs in late 2017 and early 2021), which tends to correct relatively soon as effects are cancelled out by growing total hashrate.

Sometimes, it is possible that no mining equipment is profitable during a certain period. In this case, we use the following assumption:

Assumption 2: during time periods where no mining equipment is profitable, the model uses the last known profitable equipment.

It is reasonable to assume that miners will not immediately switch off unprofitable equipment as long as the time periods are acceptably short and infrequent. 

The model applies a 14-day moving average to the profitability threshold in order to smoothen the switch from one equipment type to another as a result of short-term hashrate fluctuations and price volatility. 

The lack of available price data prior to July 18th 2010 causes our model to assume an electricity consumption of 0 given that no profitability threshold can be calculated. This means that according to our model, no rational economic actor would have mined Bitcoin before it had a price. In reality, however, some people who were not motivated by economic incentives did contribute to the network hashrate, although a simple analysis that neglects profitability altogether suggests that it comprised a daily average of less than 85 CPUs just prior to the model start – which can be safely neglected.

Constructing the lower bound estimate 

In a best-case scenario, every single miner would always use the most energy-efficient equipment that maximises expected profits. The lower bound estimates (Dlower and Elower) are thus based on the following assumption:

Assumption 3a (lower bound): all miners always run the most efficient hardware available.

This assumption also implies that miners will rapidly upgrade mining gear as soon as more energy-efficient hardware becomes available on the market. 

Power usage effectiveness (PUE) is a measure of data centre energy efficiency: data centres generally consume more energy than is required to simply run servers, mostly because of cooling, supporting IT equipment, and other overheads. The higher the ratio, the less efficiently energy is used. Data centres with PUE below 1.2 are generally considered efficient. For reference, Google’s average PUE is 1.11, whereas the average PUE of most data centres corresponds to 1.8 or more. 

In the case of Bitcoin mining, however, electricity costs account for the vast majority of operational expenditures. Mining farm operators therefore have a clear incentive to optimise cooling systems in order to reduce overall costs. Conversations with miners support the hypothesis that mining facilities generally have significantly lower PUE than traditional data centres. 

In a best-case scenario, mining facilities have optimised data centre operations to a point where there is nearly zero overhead. This scenario is represented by assuming a PUE of 1.01.

Assumption 4a (lower bound): all mining facilities have a PUE of 1.01.
The lower bound estimates can be mathematically expressed as follows:
The lower bound estimate corresponds to the absolute minimum electricity load of the Bitcoin network. While useful for providing a quantifiable floor, it is a purely hypothetical figure that is unrealistic for a variety of reasons: 
  • Not all miners use the most efficient hardware: old equipment can remain profitable for a considerable time when miners have access to cheap electricity and Bitcoin prices remain high. 

  • Long delivery and installation times: the delivery and installation of newly released equipment can take up to 3 months or longer. 
    Hardware supply shortage: the most efficient hardware may not be available in all regions in sufficient quantities. 
    Optimistic PUE: not all mining facilities have an optimal PUE. 

Constructing the upper bound estimate 

Calculating the upper bound estimate (Dupper and Eupper) is a more difficult task. 

We could imagine a worst-case scenario where every miner uses the least efficient computing device available on the market that is capable of computing cryptographic hashes a central processing unit (CPU) powering, for instance, a computer, a tablet, or even a smartphone. However, with the exponential increase of Bitcoin’s network difficulty since 2016, this assumption would quickly lead to a consumption figure that exceeds the world’s total energy production, let alone the fact that miners would need to operate at massive losses. 

We thus adjust the assumption as follows:

Assumption 3b (upper bound): all miners always use the least efficient hardware available at each time period as long as the equipment is still profitable in terms of electricity costs.

As soon as a given equipment type ceases to be profitable in electricity terms, it will be retired and replaced with the next least efficient hardware model that still remains profitable. 

It is worth remembering that the profitability threshold for each mining hardware type is calculated strictly in electricity terms and does not take into account capital expenditures nor other operational expenditures.

Assumption 4b (upper bound): all mining facilities have a PUE of 1.20.

We assume that in this scenario, all mining farms have a PUE of 1.20. While still considered efficient by general-purpose data centre standards, it ranges at the higher end of PUE figures reported by miners. 

The upper bound equations can thus be mathematically expressed as follows:

The upper bound estimate corresponds to the absolute maximum electricity load of the Bitcoin network. While useful for providing a quantifiable ceiling, it is a purely theoretical figure that is unrealistic for a variety of reasons:  
  • Miners demand the most energy-efficient hardware: large miners with industrial-scale data centres compete for gaining early access to the newest ASIC generations that are more energy-efficient. 

  • Old equipment gets replaced: many miners replace old ASIC generations that have remained unprofitable for a long time with new equipment rather than storing old equipment for years hoping for the profitability threshold to increase. 

  • Other operational expenditures have an impact, too: ignoring additional expenditures such as cooling and maintenance costs may artificially overstate the economic lifetime of inefficient hardware. 

Constructing the best-guess estimate 

Given that both the lower and upper bound estimates rely on fairly unrealistic assumptions, we attempt to provide an educated guess that more accurately quantifies Bitcoin’s real electricity usage. 

In reality, many miners do not run a single type of mining equipment, and they do not all switch to the newest-generation hardware at the same time, if they do at all. In many cases, miners operate a combination of different models as long as the equipment remains profitable in electricity terms (i.e. stay below the profitability threshold). 

The difficulty lies in determining a realistic weighting approach for all profitable equipment types on a continuous basis that takes into account changing market and network conditions over time. Analysing the market share evolution of the major mining manufacturers would be a good proxy; however, reliable market share data over multiple periods is unfortunately not available. 

In the first version of the model, we used the assumption that all miners use a weighted basket of hardware types that are profitable in electricity terms. In other words, we assumed that all profitable machines were equally distributed among miners. While there are obvious limitations to this approach (e.g. many hardware types have not been produced and sold in equal quantities, some equipment may not have been available to everyone at the same time, and other machines may already have been fully retired despite becoming profitable again for a short period of time), it has yielded surprisingly similar results when compared to other methodologies, such as for instance Stoll et al.’s (2019) market share analysis.4

Update v1.1.0 introduced a data set that estimated the market share of Bitmain Antminer S7 and S9 machines, respectively. This dynamic dataset is based on the nonce distribution analysis methodology pioneered by Coin Metrics, which has produced reliable estimates that are broadly in line with findings from alternative methodologies and studies.5 Over time, however, the methodology appears to have become less reliable as estimates tend to significantly overstate the share of operational devices. We have therefore opted to retroactively revert to an exclusively equally-weighted approach by removing the S7 and S9 parameters in v1.2.0.

We thus use the following updated assumption for our best-guess estimate:

Assumption 3c (best-guess): all miners that do not operate Antminer S7 or S9 machines are assumed to use an equally-weighted basket of hardware types that are profitable in electricity terms.

We assume that all mining farms have a PUE of 1.10 when calculating our best-guess estimate.6 This figure is slightly more conservative than other estimates but has been repeatedly confirmed during private conversations with miners and mining experts.

Assumption 4c (best-guess): all mining facilities have a PUE of 1.10.
Our best-guess estimates can thus be mathematically expressed as follows: 

Limitations of this methodology will be discussed in the next section. 


Limitations of the model 

Every model is an incomplete representation of reality that relies on specific assumptions, some of which may be debatable. As a result, every model has limitations that need to be discussed. In particular, the current CBECI model exhibits the following limitations (the list is non-exhaustive): 

  • Strong dependence on electricity cost estimate: electricity costs can significantly vary from one country, region, and provider to another. Prices are generally dynamic and adjustable, often according to seasonal circumstances, the quantity of electricity consumed, and other factors. Modifying the default electricity cost assumption can substantially change the model output. 
    Ignoring other cost factors: other potential factors that influence the decision of miners to switch off and/or replace existing equipment have not been incorporated into the model (e.g. maintenance and cooling costs). 
    Simplistic weighting of profitable hardware: assuming that all profitable equipment is equally distributed among miners is unrealistic given that not all hardware is produced in equal quantities and readily available. The exact market share is unknown, although existing data suggests that a few large manufacturers dominate the market. The lack of reliable longitudinal market share data impacts all bottom-up approaches. 

  • Hardware selection: we may not be aware of new and more efficient hardware that is not yet available on the market. Some have argued that manufacturers are using proprietary equipment to their own benefits before public release.7
    Hardware specifications may not correspond to real performance: hardware manufacturers often advertise the performance and energy efficiency of their products using best case scenarios. Furthermore, miners may decide to overclock or underclock their machines for various reasons, which the model does not take into account.  
    Short switching periods: it is unlikely that miners are able to react as quickly to short-term changes in the profitability threshold. While we attempt to smoothen the effect of short-term hashrate variations and price volatility, applying a moving average of 14 days (profitability threshold), may not be sufficient. 

While most limitations do not have a major impact on the performance of the model, we are aware of its imperfections. The CBECI is an ongoing project that is maintained on a continuous basis. The model will be refined in response to changing circumstances, with all changes being transparently highlighted in the Change Log

In case you would like to provide suggestions on how we could improve the index, please feel free to send us a message using this form.

 How does the CBECI compare to other estimates? 

There have been multiple attempts in the past to analyse the electricity consumption of the Bitcoin network and assess its environmental footprint. A list of available studies and articles is presented in Table 2. With the exception of Alex De Vries’s  “Bitcoin Energy Consumption Index” (BECI), there is no live index tracking Bitcoin’s electricity load and consumption in real time.

Table 2: Overview of previous studies


Date of publication



Stoll, C., Klaaßen, L., and Gallersdorfer, U.

June 2019

The Carbon Footprint of Bitcoin


Zade, M., Myklebost, J., Tzscheutschler, P., and Wagner, U.

March 2019

Is Bitcoin the Only Problem? A Scenario Model for the Power Demand of Blockchains


Krause, M. J., and Tolaymat, T.

November 2018

Quantification of energy and carbon costs for mining cryptocurrencies


Mora, C., Rollins, R.L., Taladay, K., Kantar, M.B., Chock, M.K., Shimada, M., and Franklin, E.C.

October 2018

Bitcoin emissions alone could push global warming above 2°C


McCook, H.

August 2018

The cost & sustainability of Bitcoin


De Vries, A.

May 2018

Bitcoin’s Growing Energy Problem


Vranken, H.

October 2017

Sustainability of bitcoin and blockchains


Bevand, M.

February 2017

Electricity consumption of Bitcoin: a market-based and technical analysis


Hayes, A. S.

March 2015

A Cost Production Model for Bitcoin


O’Dwyer, K.L., and Malone, D.

September 2014

Bitcoin Mining and its Energy Footprint


These studies tend to produce considerably diverging findings along a relatively broad range of possible estimates. This can be explained by the application of different methodologies adopted by the study authors: some use a top-down economic approach, whereas others are based on a bottom-up techno-economic approach (like the CBECI model). 

Each study relies on a set of assumptions that may be called into question. As a result, the design of each study – including our own analysis – has its own pitfalls and downsides. Some papers, however, have been criticised for applying overly simplistic assumptions and containing non-trivial errors such as inappropriate averaging over time periods or simple extrapolations. For a more thorough review of previous studies, see Koomey (2019).8

The CBECI has been designed with the aforementioned studies in mind. We have carefully reviewed the various methodologies and incorporated best practices. This website attempts to provide comprehensive documentation with transparent version control, highlight the model’s dependence on the electricity cost assumption by allowing visitors to adjust the default value, and openly present the uncertainties and limitations of the model. Feedback and suggestions for further improvements can be given here.