How much energy does Bitcoin use? How it is measured, and what actually drives it

Nobody meters Bitcoin’s electricity, so every number you have seen is a model

There is no meter on Bitcoin. Miners are private businesses spread across the world, most do not publish their power bills, and the network itself records nothing about electricity. So every figure you have seen is the output of a model, and the most cited model is the Cambridge Bitcoin Electricity Consumption Index, run by the Cambridge Centre for Alternative Finance at the University of Cambridge.

The index works backwards from what can be observed. The network’s hash rate is estimated from difficulty and observed block production; individual mining attempts are not published. The efficiency of each machine model, in joules per terahash, is published by manufacturers and adjusted by Cambridge’s experts. The model then assumes miners are rational and run a machine only while it is profitable, using a default electricity price of $0.05 per kilowatt-hour to decide which machines qualify. From that it publishes three numbers: a lower bound assuming everyone uses the most efficient hardware available, an upper bound assuming the least efficient hardware still worth running, and a best guess built from a weighted basket of models, with more weight on newer machines and an allowance for shipping and installation time.

Cambridge is candid about the limits. Changing the electricity-price assumption “can substantially change the model output.” The global mix of hardware “cannot be precisely determined.” Manufacturers advertise best-case efficiency. And the methodology page includes a line worth keeping in mind for every number below: “Every theoretical model is an incomplete representation of reality.” Its survey of other published estimates found a spread from under 1 to over 200 terawatt-hours a year for the same network, which tells you how much the assumptions matter.

What drives it: hash rate and hardware, not the number of transactions

Bitcoin’s electricity use is a function of two things: how much hashing is happening and how efficiently. A 2021 paper in Nature Communications states the arithmetic directly, computing energy as the network hash rate multiplied by the mining hardware’s efficiency and by a factor for the facility’s cooling and overheads. Our article on how mining works explains why the hashing exists at all; the article on hash rate explains how the figure is estimated from the chain.

What sets the hash rate is money. Mining pays a reward in bitcoin per block, so when the price rises, more machines become profitable and get switched on; when it falls, or power gets expensive, machines are switched off. The White House Office of Science and Technology Policy captured the long-run result in its 2022 report: between August 2016 and July 2022 the energy used per unit of hashing fell by about 85% as hardware improved, but hash rate rose by more than 14,000%, so estimated electricity use still rose by about 2,000%. Efficiency gains were “negated by rising hashrates as mining competition has increased.”

A miner hashes a fixed-size block header rather than doing one proof-of-work search per payment. Dividing estimated network energy by a transaction count is an allocation ratio, not a measurement of the extra electricity caused by one more payment. Demand and fee revenue can nevertheless affect mining profitability indirectly. The reward per block also shrinks on a fixed schedule, and the OSTP report notes that future reductions to mining rewards “may limit the growth of Bitcoin electricity usage.” Our dossier on halvings and supply covers that schedule.

The headline numbers, and the dates they belong to

With those caveats, here are the numbers the main institutions have published, each tied to its date. The OSTP report of September 2022, written under Executive Order 14067 of 9 March 2022, put global electricity use for all crypto-assets at 120 to 240 billion kilowatt-hours a year as of August 2022, or 0.4% to 0.9% of world electricity, a range it called comparable to all conventional data centers in the world. It attributed 60% to 77% of that to Bitcoin, or 90 to 145 billion kilowatt-hours. For the United States, which it said hosted about 38% of Bitcoin hash rate that August, it estimated 0.9% to 1.7% of national electricity use for proof-of-work mining and 25 to 50 million metric tons of carbon dioxide a year, similar to diesel burned by American railroads.

The U.S. Energy Information Administration followed on 1 February 2024 with a preliminary estimate that all cryptocurrency mining used 0.6% to 2.3% of U.S. electricity, or 25 to 91 terawatt-hours a year. It had identified 137 mining facilities and gathered capacity data on 101 of them, totaling 10,275 megawatts, and it warned that the bounds were “quite wide,” that facilities often run below capacity, and that operations “can be moved rapidly to areas with lower electricity prices.”

The Cambridge survey-based estimate cited here, published 28 April 2025, put the network at 138 terawatt-hours a year, about 0.5% of global electricity, with emissions of 39.8 million metric tons of carbon dioxide equivalent. That figure rests on responses from 49 mining firms that together represent about 48% of global mining activity. All three are estimates with dates attached; none of them is the number today.

The debate about where the power comes from

How much electricity is one question; where it comes from is a separate one, and the more contested. The Cambridge 2025 survey found 52.4% of respondents’ power came from sustainable sources, counting 9.8% nuclear and 42.6% renewables such as hydro and wind, up from 37.6% in its 2022 estimate. Natural gas had become the largest single source at 38.2%, up from 25.0%, while coal fell to 8.9% from 36.6%. Those are self-reported figures from firms that chose to answer, which the researchers present as the best available evidence rather than a census.

The fall in coal has a specific cause. Until 2021 a large share of mining ran in China, partly on coal and partly on seasonal hydro, and the 2021 mining ban pushed most of it out. The Nature Communications paper mentioned above had projected, in April 2021, that Chinese mining would peak in 2024 at 296.59 terawatt-hours and 130.50 million metric tons of carbon dioxide if nothing changed. Two months later, everything changed. It is a useful reminder that projections in this field go stale fast.

The argument miners make is that they can use energy nobody else wants. Cambridge’s comparisons page describes “surplus energy that would otherwise have been lost or wasted,” such as curtailed renewables and flared gas, and notes that mining’s mobility lets it go where that energy is. The OSTP report takes the methane case seriously: venting and flaring wastes about 4% of global methane production, and mining that burns methane which would otherwise be vented “can yield positive results for the climate.” It also notes that replacing an existing flare does not change emissions, and that capturing the gas with conventional equipment is an alternative. On the grid side, the same report flagged Texas, where mining used about 3% of peak demand and could add as much as 25 gigawatts over a decade, as a test of whether flexible demand helps or strains reliability.

How to read any Bitcoin energy claim

Three questions sort most claims. Which date? Global crypto-asset consumption rose by more than 67% between July 2021 and January 2022 and then fell 17% by August 2022, according to OSTP, so an undated figure is close to meaningless. Which bound? Cambridge’s lower and upper estimates can differ by a factor of several, and a headline that quotes one without saying so is choosing a story. Compared with what? Cambridge warns that country comparisons “provide only limited insight” because every country’s energy profile is unique, and that its gold-mining comparison rests on 2006 data.

And who made it? The estimates above come from a university research center, a White House science office, a federal statistics agency and a peer-reviewed journal, and they disagree with each other by tens of terawatt-hours while agreeing on the shape: something under 1% of world electricity, driven by price and hardware, trending toward gas and renewables and away from coal, and uncertain at the edges. A claim more precise than that is more precise than the evidence.