How Researchers Estimate Bitcoin’s Energy Use

Bitcoin miners use specialized computers to compete in proof of work, the calculation that secures the network and adds blocks to its public record. Because no single meter measures all miners’ electricity use, researchers estimate it from network computing activity and assumptions about mining machines and facilities.

Gigawatts measure estimated power demand at a point in time or over an averaging period. Terawatt-hours per year show how much electricity that demand would use if it continued for a year; they do not automatically measure Bitcoin’s electricity use during a completed year.

Estimate electricity use from hashrate

Researchers estimate Bitcoin’s hashrate, the total computing rate miners contribute to the network, using network data. They then estimate how much electricity the hardware needs to produce that hashrate.

The basic relationship is:

Estimated mining electricity demand ≈ network hashrate × hardware energy use per unit of hashrate

Hardware energy efficiency is commonly measured in joules per terahash (J/TH). A lower figure means a machine uses less energy to perform the same amount of hashing. Researchers also adjust for facility overhead when their estimate includes electricity for cooling and other supporting equipment.

The calculation depends on knowing what hardware miners operate. That inventory is not public or complete. Researchers therefore estimate the mix of machines, their operating conditions, and how much electricity facilities use beyond the mining computers.

How CBECI estimates Bitcoin electricity use

The Cambridge Centre for Alternative Finance’s Bitcoin Electricity Consumption Index, or CBECI, estimates electricity demand by modeling mining hardware. Its catalog includes more than 100 ASIC models. ASICs, or application-specific integrated circuits, are computers designed for a particular task, such as Bitcoin mining.

The model assumes miners run machines when mining revenue covers their costs. By default, it uses an assumed global average electricity price of $0.05 per kilowatt-hour to estimate which hardware would remain profitable. Real electricity prices vary by location and contract, so this assumption affects which machines the model counts as active.

CBECI publishes a lower bound, a best guess, and an upper bound. The lower bound assumes miners use the most efficient hardware that remains profitable and have little facility overhead. The upper bound assumes less efficient profitable hardware and more overhead. The best guess uses a weighted mix of hardware. These are modeled scenarios, not a statistical confidence interval.

Facility overhead is represented through power usage effectiveness, or PUE. PUE compares a facility’s total electricity use with the electricity used by its computing equipment. CBECI’s current methodology uses PUE assumptions of 1.01 for its lower bound, 1.10 for its best guess, and 1.20 for its upper bound.

Facility, revenue, and survey estimates

A facility-based estimate starts with identified mining sites and details such as their equipment, power capacity, or electricity use. It can use information from actual operations, but a count of known facilities may miss sites or mistake their maximum capacity for their typical use.

In a 2024 analysis, the U.S. Energy Information Administration (EIA) estimated U.S. Bitcoin mining at 25 to 91 terawatt-hours (TWh) for 2023. That estimate started with Cambridge’s global figures and assumed the United States accounted for about 38% of global mining activity. In a separate facility-based calculation, EIA identified 137 U.S. cryptocurrency-mining facilities. The 101 facilities with known maximum electricity capacity totaled 10.275 gigawatts. At an assumed 80% utilization, that capacity would amount to about 70 TWh per year. The facility figure covered cryptocurrency mining generally, so it should not be treated as a Bitcoin-only estimate.

A revenue-based estimate works backward from mining income. In simplified form, it assumes what share of miners’ revenue goes to electricity, then divides that spending by an assumed electricity price. One historical example used a 60% share and an electricity price of $0.05 per kilowatt-hour. The result changes when either assumption changes.

Researchers also use surveys of mining firms. Cambridge’s 2025 industry report used survey responses to estimate weighted hardware efficiency at 28.2 J/TH. For June 30, 2024, it estimated annualized Bitcoin electricity use at 138.2 TWh, close to CBECI’s modeled estimate of 137.4 TWh for the same date. The hardware-efficiency comparison had 45 survey responses. The close estimates offer a useful comparison between methods, but the survey does not represent a direct measurement of every miner.

Why Bitcoin electricity estimates differ

The hardware mix is unknown: newer, more efficient machines use less electricity to produce a given hashrate than older machines. Researchers have to estimate how many of each type remain in service, and reliable historical market-share data are limited.

Electricity prices also shape estimates. Miners face different rates and contract terms, while a global model needs a simplified assumption. In Cambridge’s model, the assumed price helps determine which hardware remains profitable and therefore affects estimated power demand.

Each estimate also covers a specific place, activity, and time. A global Bitcoin estimate covers different activity from a U.S. estimate or a count of all cryptocurrency-mining facilities. The network’s hashrate changes over time. Cambridge averages the hashrate over seven days for its annualized figure to smooth short-term changes, so a reported value depends partly on the date and averaging window.

Each method also has gaps. Models can estimate activity across the whole network, but they rely on assumptions about machines, costs, and facilities. Facility counts and surveys use real-world information, but they can miss operations or reflect an incomplete sample. Compare estimates with those differences in mind.

How to read annualized Bitcoin energy figures

An annualized consumption figure shows how much electricity an estimated power demand would use over a year if it stayed constant. For example, 19 gigawatts sustained for a year would equal about 166 TWh. EIA reported Cambridge’s best estimate of 19.0 GW at the end of January 2024, equivalent to about 170 TWh per year after rounding. The reported theoretical range was 9.1 to 44.0 GW, or 80 to 390 TWh per year.

Those annualized values describe a rate translated into a yearly equivalent. They do not show that Bitcoin used 170 TWh during the preceding calendar year. To estimate electricity use over a past period, researchers need to accumulate estimated consumption across that period. Cambridge, for example, calculates monthly consumption by summing daily estimates.

The dates help explain why several published numbers can all be valid. EIA cited an estimate of 120 TWh, with a range of 67 to 240 TWh, for global Bitcoin mining during 2023. Cambridge estimated 137.4 TWh per year on June 30, 2024, and 183 TWh per year on December 31, 2024. The latter is an annualized endpoint, not a total for calendar 2024. These figures describe different periods and quantities.

When comparing estimates, check three details first: the date, the geographic and activity scope, and whether the unit is GW, TWh per year, or TWh over a completed period. A figure that omits those details is easy to misread.

Electricity is not emissions

Electricity use alone does not determine Bitcoin’s greenhouse-gas emissions. Estimating emissions also requires information about the electricity sources used and their emissions rates. A power-demand estimate therefore cannot, by itself, establish the network’s carbon footprint.

Energy-per-transaction comparisons need similar care. Bitcoin mining’s electricity demand is tied to the operation of proof of work, rather than being a simple fixed amount of electricity for each payment. Dividing estimated network energy by transaction count does not directly measure the energy caused by one transaction.