Bitcoin energy consumption is set by the machines that mine it, and their number moves fast. At 10:06 UTC on 9 November 2026 they were computing roughly 963 exahashes per second, according to mempool.space’s three-day estimate. Hashrate Index recorded a seven-day peak of 1,156 EH/s on 18 October 2025; by June 2026 Texas curtailment had pulled the network down to 853 EH/s. The electricity draw moved roughly in step, by about a quarter in eight months, while the number of transactions barely mattered. The power goes into security, and it follows the price of bitcoin and the reward for blockchain and earns a reward, which makes rewriting history very expensive.">mining it, not the traffic.
Two questions get mixed up in that argument. The first is a measurement: how many terawatt-hours a year does the network use? The answer depends on who estimates and when, and the serious estimates range from 138 to about 204 TWh. The second is a judgement: is that a problem? That depends on where the power comes from, what else it could have done, and what you compare it with. This lesson gives the numbers with their dates and methods first, then each side’s strongest case, then how governments have reacted.
A short reminder before going deeper. In proof of work, miners repeat a hash computation trillions of times a second until one finds a block, and the network adjusts the difficulty every 2,016 blocks so that a block arrives about every ten minutes. The mechanism is explained in our lesson on proof of work in depth, and the link between hashrate and difficulty in the hashrate and difficulty lesson. For energy, one consequence matters: more machines do not produce more blocks or more transactions, they only make each block harder to find. Electricity is the cost of that difficulty.
Energy is also quoted in two units that are easy to confuse. Power is a rate, in gigawatts (GW): what the machines draw at a given moment. Energy is power over time, in terawatt-hours (TWh): one gigawatt running for a full year is 8.76 TWh. So an estimate of 165 TWh a year means an average draw of about 18.8 GW, and 138 TWh means about 15.8 GW.
Bitcoin energy consumption according to the Cambridge index
The most quoted figure comes from the University of Cambridge. Its Cambridge Bitcoin Electricity Consumption Index (CBECI), run by the Cambridge Centre for Alternative Finance, publishes a daily estimate of the network’s power demand and turns it into an annualised consumption. The index was the source behind the figures in our earlier lessons, and its method explains why its number moves.
How the CBECI builds its estimate
The CBECI methodology page calls its approach a “hybrid top-down” model. It starts from the network hashrate, which anyone can infer from the block record. It then asks which machines could be producing that hashrate. Cambridge keeps a basket of more than 100 mining devices, each with its efficiency in joules per terahash (J/TH), and assumes miners are rational: they run only hardware that is profitable at the electricity price assumed.
That price is fixed. The page states that the model assumes a global average of 0.05 USD per kWh, “constant over time”, while admitting that real prices vary widely between and within countries. Visitors to the index’s landing page can choose a different electricity cost to see how the estimate reacts.
From there the index gives three readings. The theoretical lower bound assumes every miner uses the most efficient machine on the market, in a facility with a power usage effectiveness (PUE) of 1.01, meaning almost no overhead for cooling. The upper bound assumes the least efficient machine that is still profitable, at a PUE of 1.20. The best guess sits in between: a weighted basket of profitable machines, with older models weighing less as they age, and a PUE of 1.10 for every farm.
Three details make the index lag the real network. A new machine is only counted two months after its release, to allow for shipping and installation. The profitability filter uses a 14-day moving average, so unprofitable hardware is not switched off in the model at once. And the annualised consumption is itself smoothed with a seven-day moving average, to damp short swings in hashrate. None of this is hidden; it is simply why the CBECI describes the network of a few weeks or months ago rather than today’s.
The latest reading: about 165 TWh, with data ending in May 2026
Read on 7 November 2026, for our guide to Bitcoin mining, the CBECI showed a best-guess estimate of about 165 TWh a year. The series displayed that day ended in early May 2026, so the figure describes the network of last spring, when hashrate was lower than in November. Any CBECI number quoted without the date its data end is incomplete.
The Cambridge industry report of April 2025: 138 TWh
The second Cambridge figure comes from a different exercise. For the Cambridge Digital Mining Industry Report, published in April 2025, the centre surveyed 49 mining companies, listed and private, headquartered in 16 jurisdictions and operating in 23 countries. Together they represented about 48 percent of Bitcoin’s hashrate, some 268 EH/s when the data was collected.
From those answers the report estimated the network’s electricity use at 138 TWh a year, up 17 percent on the year before and about 0.54 percent of world electricity. It also found that average hardware efficiency had improved by 24 percent in a year, to about 28.2 J/TH by June 2024, that electricity made up more than 80 percent of miners’ cash operating costs, and that the median price paid was about $45 per megawatt-hour for electricity alone.
The 138 TWh figure is honest about its date, and it is now old. If 268 EH/s was 48 percent of the network, the whole network was running at about 558 EH/s. On 9 November 2026 it was running at about 963 EH/s, roughly 73 percent more. Machines have become more efficient since, so consumption has not risen by the same proportion, but a figure from a network that much smaller cannot describe today’s.
Why 138, 165 and 204 TWh are not contradictory
A third widely quoted number comes from Digiconomist, the site founded by researcher Alex de Vries, whose peer-reviewed work on bitcoin’s footprint is cited further down. Its Bitcoin Energy Consumption Index, read on 9 November 2026 (the page carries no update date), gave 204.44 TWh a year. Its method is top-down and economic: it takes miners’ annualised income, about $12.5 billion, estimates that 81.72 percent of it goes on electricity at 5 cents per kWh, and converts that spending into energy. Its carbon figures rest on a grid mix measured in August 2021.
| Source | Method | Annual estimate | Date |
|---|---|---|---|
| Cambridge Digital Mining Industry Report | Survey of 49 firms (about 48% of hashrate), scaled up | 138 TWh | April 2025 |
| CBECI best guess (Cambridge) | Hashrate plus a basket of profitable machines, 0.05 USD/kWh | About 165 TWh | Read 7 Nov 2026, data ending early May 2026 |
| Digiconomist | Miner income converted into electricity spending | 204.44 TWh | Read 9 Nov 2026, page undated |
| ethereum.org comparison table (Cambridge data) | Quoted figure | 149 TWh | Sources accessed July 2023 |
| Order of magnitude (Coindaro calculation) | 963 EH/s at 17 to 20 J/TH, before cooling overhead | About 143 to 169 TWh | 9 Nov 2026 |
The last row is our own arithmetic, labelled as such, to show how the numbers fit. Multiply 963 EH/s by an assumed fleet average of 17 to 20 J/TH and you get a draw of 16.4 to 19.3 GW, or 143 to 169 TWh a year before cooling and other overhead. The assumption is ours: Hashrate Index’s monthly review sorts machines into tiers of under 14, 14 to 19, 19 to 25 and 25 to 38 J/TH, and the real average depends on how many old machines are still switched on. The spread between 138 and 204 TWh comes from three choices: the date of the hashrate used, the efficiency assumed for the fleet, and whether one starts from machines (Cambridge) or from money (Digiconomist).
Where the power comes from: renewables, gas and nuclear
The total says nothing about harm on its own. A terawatt-hour from a hydro dam in a remote valley and one from a coal plant next to a city are not the same thing for the climate or for the people breathing nearby. That is why the energy mix is the most disputed number in the debate.
What the Cambridge survey found
The April 2025 report asked its respondents where their electricity came from. Its answer, compared with the previous survey, shows a large shift away from coal and towards gas:
| Source of electricity (Cambridge survey) | 2024 | 2022 |
|---|---|---|
| Renewables (hydro, wind, solar and others) | 42.6% | not split in the report summary |
| Nuclear | 9.8% | not split in the report summary |
| Sustainable total (renewables plus nuclear) | 52.4% | 37.6% |
| Natural gas | 38.2% | 25.0% |
| Coal | 8.9% | 36.6% |
Read the table with care. “Sustainable” in the report includes nuclear, so the renewable share is 42.6 percent, not 52.4. The survey covers firms representing about 48 percent of hashrate, and the respondents’ operations were 75.4 percent in the United States, a share the report itself calls likely overstated because American firms answered in large numbers. And the figures are self-reported by companies with an interest in the answer. The report also estimated emissions at 39.8 million tonnes of CO2 equivalent a year, about 0.08 percent of the world total, while noting that a model based on the location of mining IP addresses gives 69.6 million tonnes.
Counter-evidence from outside the industry
Studies that do not rely on miners’ own answers tend to find a dirtier mix. In a paper in Nature Communications published on 26 March 2025, a Harvard-led team located 34 of the largest US mines and traced which power plants responded to their demand. From mid-2022 to mid-2023, the authors write, those mines used 32.3 TWh, 33 percent more than Los Angeles, and 85 percent of it came from fossil fuels. They estimated that 1.9 million Americans were exposed to at least 0.1 microgram per cubic metre of extra fine-particle pollution as a result.
That study has its own critics. As reported by The Block on 17 April 2025, Daniel Batten, a climate-tech investor who co-founded CH4 Capital and sits on the advisory board of the miner MARA, called it deeply flawed: in his view it applied marginal emissions modelling from WattTime’s proprietary tool without stating its limits, picked its sites selectively and leaned on news reports. The Digital Assets Research Institute, which describes itself as independent, published a rebuttal along similar lines. The disagreement is partly about method: marginal accounting asks which plant ramps up when a new load arrives, average accounting looks at the whole grid’s mix, and the two can give very different answers for the same mine.
Flared gas: burning it for hashes instead of for nothing
Oil wells often produce gas that producers cannot sell, because no pipeline reaches them. Much of it is flared, burned on site. The World Bank’s Global Gas Flaring Tracker, updated in June 2026, counted 167 billion cubic metres flared worldwide in 2025. Some miners put generators at the wellhead and run machines on that gas, arguing that burning it in an engine releases less methane, a far stronger greenhouse gas than CO2, than an imperfect flare.
The best-documented case is in North Dakota. CNBC reported on 26 March 2022, citing people familiar with the project, that ExxonMobil had run a pilot with Crusoe Energy Systems in the Bakken region since late January 2021 and expanded it in July 2021, feeding gas that would otherwise be flared to bitcoin miners. Exxon did not comment. Crusoe’s president told CNBC that more than 10 million cubic feet of gas a day was no longer flared thanks to its systems, and the company claimed about 63 percent lower emissions than continued flaring. Those are the company’s own figures.
The Cambridge report counts flared-gas use as one of the measures that can lower its emission estimates, but neither it nor the other estimates in this lesson measure how much of the network runs on such gas. Critics add a second objection: if mining makes stranded gas profitable, it may keep marginal wells running longer than they otherwise would. The scale of flared-gas mining is therefore unknown, and both its benefit and its cost are argued from cases, not from totals.
Flexible load: Texas and the grid
Miners have one property most industrial users lack: they can switch off in minutes and lose only the revenue of those minutes. The Cambridge report says respondents curtailed 888 GWh of load in 2023. In Texas, on the ERCOT grid, miners cut their use during peak hours, both to sell power back at high prices and to lower their transmission charges. Hashrate Index described the effect in 2026: during the late-January Arctic blast the network’s seven-day average fell from 1,051 to 833 EH/s in one week, and through the summer, as Texas miners avoided the four coincident peak (4CP) hours that set their transmission charges, hashrate kept dipping below 900 EH/s.
Supporters call this a grid service: a buyer of last resort that absorbs surplus wind and solar and steps aside when households need the power. Critics answer that a flexible load is still a load, that payments for curtailment are a cost to other ratepayers, and that new demand can delay the retirement of fossil plants. Texas lawmakers have since legislated on large loads in general: Senate Bill 6, signed on 20 June 2025, deals with planning, interconnection and costs for large electrical loads.
Ethereum’s energy use fell by more than 99.9% after The Merge
Ethereum offers the one real-world before-and-after. On 15 September 2022, The Merge replaced its proof of work with proof of stake: instead of miners burning electricity, validators lock ether as collateral and lose part of it if they cheat. The ethereum.org energy page cites the Crypto Carbon Ratings Institute (CCRI), which estimated that the switch cut Ethereum’s annual electricity use by more than 99.988 percent, to about 2,601 MWh, and its carbon footprint by about 99.992 percent, from 11,016,000 to 870 tonnes of CO2 equivalent. The roadmap page of the same site rounds it to about 99.95 percent. Both are estimates, but no one disputes the order of magnitude.
Why does Bitcoin not do the same? Because its security model is the point of contention. Supporters of proof of work argue that energy spent is a cost an attacker cannot fake or recover, while staking, they say, ties control to existing holdings; supporters of proof of stake answer that economic penalties do the same job for a tiny share of the energy. Our comparison of consensus mechanisms sets out both views, and our introduction to Ethereum covers what changed for its users. A change of that size to Bitcoin would need broad agreement among developers, node operators, miners and users, and our lesson on Bitcoin governance and BIPs shows how hard that is even for small changes.
Banking, gold and data centres: what the comparisons show and where they fail
Raw terawatt-hours mean little to most readers, so both sides compare. Countries are the classic yardstick; banking and gold are the favourites of Bitcoin’s defenders, because bitcoin is pitched as both a payment system and a store of value. Each comparison tells you something and hides something.
Galaxy Digital’s 2021 comparison with banking and gold
The comparison most cited by the industry came from Galaxy Digital Mining in May 2021, in a report titled On Bitcoin’s Energy Consumption: A Quantitative Approach to a Subjective Question. As reported by Decrypt on 17 May 2021, it put Bitcoin at about 113 TWh a year, the banking system at about 263 TWh and the gold industry at about 240 TWh. According to Decrypt, the banking figure counted bank data centres, branches, ATMs and card networks (central banks excluded), and the gold figure was obtained by converting estimates of the industry’s greenhouse-gas emissions into energy with an International Energy Agency converter.
The numbers were disputed from the start. Decrypt noted that Cambridge’s index then put Bitcoin at about 144 TWh, well above Galaxy’s figure. More basically, the banking total buys services Bitcoin does not provide: branches, loans, deposit insurance, payroll, card payments for billions of people. Bitcoin would not replace most of that energy even if it were widely adopted, and the report came from Galaxy’s own mining business.
A neutral reference point exists. The comparison table on the ethereum.org energy page, built from sources accessed in July 2023, lists Bitcoin at 149 TWh, gold mining at 131 TWh and global data centres at 190 TWh. On those figures Bitcoin and gold mining are of the same order, which is the defenders’ point; the critics’ point is that gold mining produces a physical metal used in jewellery and industry, and that two large footprints do not cancel each other.
Square and ARK: the renewable-buyer argument
In April 2021 Square (now Block) published a white paper, Bitcoin Is Key To An Abundant, Clean Energy Future, with a section by ARK Invest. Its argument was that miners are flexible, location-agnostic buyers who could make more solar and wind projects profitable by consuming power that would otherwise be curtailed. As summarised by Bitcoin Magazine on 21 April 2021, ARK’s model suggested renewables could meet about 40 percent of a grid’s needs without mining and about 99 percent with mining, solar and batteries combined. Those are modelled scenarios from two firms with a stake in bitcoin, not observations of real grids.
Data centres and AI: a newer yardstick
The comparison that has grown fastest since 2024 is with data centres. The International Energy Agency estimated in its Energy and AI report that data centres used about 415 TWh in 2024, around 1.5 percent of world electricity, and projected about 945 TWh by 2030. Set against that 2024 figure, Bitcoin’s 138 to 204 TWh equals a third to a half of all data-centre use, from a single application. The same comparison cuts the other way too: miners and AI companies now compete for the same grid connections.
Why energy per transaction is a weak measure
Digiconomist also divides its total by the number of transactions and gets about 806 kWh per Bitcoin transaction, as read on 9 November 2026. The division is correct; what it implies is not. A block costs the same energy whether it holds a few hundred transactions or several thousand; an empty network would burn the same power as long as the reward paid for it. One on-chain transaction can also settle many payments, as exchange withdrawals in batches and the channels of the Lightning Network do. Defenders therefore reject the metric, and the critics’ better reply is to compare totals with what the network delivers, which is a judgement, not a measurement.
Emissions, e-waste and the machines themselves
Electricity is the headline, but it is not the only footprint. Emissions depend on the mix discussed above, and the hardware itself becomes waste quickly, because a machine that cannot pay its power bill is worthless whatever its physical condition.
| Measure | Estimate | Source and date |
|---|---|---|
| Emissions | 39.8 Mt CO2e a year (0.08% of world) | Cambridge survey, April 2025 |
| Emissions | 69.6 Mt CO2e a year | IP-location model cited in the same Cambridge report |
| Emissions | 114.03 Mt CO2 a year | Digiconomist, read 9 Nov 2026 (carbon intensity from August 2021 data) |
| E-waste | 30.7 kilotonnes a year | de Vries and Stoll, as of May 2021 |
| E-waste | 22.83 kilotonnes a year | Digiconomist, read 9 Nov 2026 |
| E-waste | About 2.3 kilotonnes in 2024 | Cambridge survey, April 2025 (86.9% of retired hardware resold, reused or recycled) |
The e-waste gap is the widest of all, more than tenfold. Alex de Vries and Christian Stoll, in a 2021 paper in Resources, Conservation and Recycling, assumed machines are discarded once unprofitable and estimated 30.7 kilotonnes a year as of May 2021. The Cambridge survey asked miners what happened to their retired machines and found that most were resold or recycled, giving about 2.3 kilotonnes. Both may be right about what they measure: a machine resold to a miner with cheaper power is not waste yet, but it will be one day.
Economists have also tried to price the damage. In a study in Scientific Reports published on 29 September 2022, Benjamin Jones, Andrew Goodkind and Robert Berrens estimated that from 2016 to 2021 each dollar of bitcoin market value created came with 35 cents of global climate damage, a ratio they place between beef production and gasoline. Their estimate uses the energy mix of that period, before the shift away from coal reported by Cambridge.
Efficiency keeps improving, which slows the growth of consumption without stopping it. Hashrate Index’s October 2026 review gave an average industrial power cost of $48 per megawatt-hour and showed what one megawatt-hour earned in each machine tier: about $132 for machines under 14 J/TH but about $49 for those at 25 to 38 J/TH. The oldest fleets are the first to be switched off when revenue falls.
What will move the number next
Bitcoin’s energy use is set by what mining pays. Revenue comes from the block subsidy, cut in half every 210,000 blocks as explained in our lesson on the halving, and from fees, which in a Cambridge study of 20 October 2026 made up only 0.67 percent of miner revenue over the 90 days to 4 October. When the subsidy halves and the price does not follow, the least efficient machines stop and consumption falls; when the price rises, machines return.
A second force is competition for power. Several listed miners are moving their sites and grid connections to artificial-intelligence computing: in early November 2026, IREN and Core Scientific set year-end dates to leave bitcoin mining for AI. If others follow, some of the electricity counted in Bitcoin’s total will be counted in data centres’ instead, without changing the grid’s load.
Policy: how governments have responded
In the European Union there is no ban on proof of work. When the European Parliament’s economic committee adopted its MiCA position in March 2022, it included no ban and asked instead for the Commission to propose bringing crypto mining with a substantial climate impact into the EU’s sustainable-finance taxonomy. What MiCA does require is disclosure. Commission Delegated Regulation (EU) 2025/422, adopted on 17 December 2024 and published in the Official Journal on 31 March 2025, sets the sustainability indicators that white papers and crypto-asset service providers must publish about a consensus mechanism. Total energy consumption in kilowatt-hours per year (field S.8) is mandatory. Where it exceeds 500,000 kWh, as it does for Bitcoin, supplementary indicators follow: the renewable share, energy per validated transaction, scope 1 and scope 2 greenhouse-gas emissions and emissions per transaction. Our MiCA guide explains the wider framework.
In the United States, the binding limits on mining’s energy use have come from states; the SEC and the CFTC regulate markets, not electricity. On 22 November 2022 New York’s governor, Kathy Hochul, signed a two-year moratorium on new air permits, and on renewals that would raise mining’s power use, for fossil-fuel plants that supply proof-of-work mining behind the meter. It also ordered a generic environmental impact statement from the state’s Department of Environmental Conservation. Under the bill text, mining on power that emits no greenhouse gases falls outside its scope.
At federal level, the Energy Information Administration tried to measure the industry directly. Its analysis of 1 February 2024 put US mining at 0.6 to 2.3 percent of the country’s electricity and launched an emergency survey of miners. The Texas Blockchain Council and Riot Platforms sued; a federal court in Texas granted a temporary restraining order on 23 February 2024, and under a settlement of 1 March the EIA dropped the emergency collection and agreed to destroy the data received. The EIA page now says the collection has been discontinued.
Is Bitcoin bad for the planet? The strongest case on each side
The critics’ case, as made by Digiconomist and by the academic studies above, runs like this. Bitcoin’s estimated 165 TWh a year is more than a third of France’s entire electricity consumption, 446.2 TWh in 2025 according to grid operator RTE, for a payment network whose capacity is fixed by its block size. Its consumption, they argue, rises with its price, with no cap built in. Independent studies of US mines find mostly fossil power and measurable air pollution, and every machine ends as waste. Proof of stake shows the same kind of network can run on a tiny fraction of the energy.
The defenders’ case, as made by miners, Cambridge’s survey respondents and pro-Bitcoin researchers, runs differently. The energy buys a security budget that no one can counterfeit. Mining is unusually flexible and mobile, so it can use stranded, curtailed or flared energy and can switch off when grids are stressed. The surveyed mix has moved from coal to gas and renewables. And the comparisons with banking, gold and data centres show other industries using as much or more.
What the data supports is narrower than either side’s slogans. The scale is real and dated: 138 TWh in April 2025, about 165 TWh in the CBECI series to May 2026, 204 TWh on Digiconomist’s method. The mix is contested, with self-reported surveys more favourable than outside studies. And whether the energy is worth it depends on the value one places on what Bitcoin does, which no index can measure.
How to read the next headline about Bitcoin and energy
A few questions sort out most claims. Which source is it, and is it a survey, a machine-based model or an economic model? What date does the underlying data end on, not the date of the article? Is the figure a total in TWh, a draw in GW, an emission in tonnes or a per-transaction ratio? Does a renewable share include nuclear? Does a country comparison use that country’s current consumption or an old one?
Then check the network itself. Hashrate is public and updated every block; if it has risen by half since the study was made, so has the work the machines do, even if efficiency has absorbed part of it. The next difficulty retarget, which mempool.space expected on 9 November to come around 13 November 2026 with a cut of about 6 percent, will show whether this autumn’s dip in revenue has switched machines off.
