On 21 December 2022, mempool.space’s daily estimate put the bitcoin hashrate at 245 exahashes per second (EH/s). On Christmas Day the same series showed 156 EH/s. Four days later it read 299 EH/s. The network did not lose a third of its machines and then double in size within a week. Some machines really were switched off during a storm in Texas, but most of that swing was the estimate itself wobbling, because nobody measures hashrate. It is inferred, after the fact, from how quickly blocks arrive.
Difficulty is the opposite kind of number. It is written into every block header, every node computes it from the same rules, and on 8 November 2026 at 09:59 UTC it stood at exactly 148,364,561,003,867, set at block 973,728 on 29 October. Against that exact figure, mempool.space estimated the hashrate over the previous three days at about 1,000 EH/s, roughly one zettahash per second. The word “about” carries a lot of weight in that sentence.
The machines, the pools and hashprice (the miner’s daily revenue per unit of hashrate) are covered in the Bitcoin mining guide. The network as a whole is a different story: how its computing power is estimated, how the difficulty adjustment chases it, which events have moved both, and how other blockchain and earns a reward, which makes rewriting history very expensive.">proof-of-work chains built different repairs for the same problem.
Difficulty adjustment in one paragraph: the 2,016-block rule
Our lesson on proof of work in depth takes the mechanism apart, from the target and nBits to the off-by-one and the timewarp attack. The short version: every 2,016 blocks, each node compares how long the period took with two weeks and scales the target accordingly, by at most a factor of four either way. Blocks came too fast, difficulty rises; too slow, it falls. Nobody votes and nobody announces it. The one formula worth carrying into this lesson links the two numbers: at difficulty D, finding a block takes on average D times 2 to the power 32 hashes, so the hashrate that would produce exactly one block every 600 seconds is D x 232 / 600. With 8 November’s difficulty, that is about 1,062 EH/s (our arithmetic).
Read that formula backwards and you have the whole method of estimating hashrate. Nodes know D. They see when blocks arrive. If blocks come every 600 seconds, the network is doing about 1,062 EH/s; if they average 630 seconds, as they did in the period that began on 29 October 2026, it is doing about 5% less, close to 1,011 EH/s. Everything else is a question of how many blocks you average over.
How the bitcoin hashrate is estimated from blocks
Miners do not report their hashrate to the network, and there is no message in the protocol that would let them. A smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool can publish what its own customers submit, but nobody outside can check it. What every node can check is the chain: each block’s difficulty, its timestamp and the cumulative work behind it. Jameson Lopp, who has written the clearest public analysis of the problem, compares it to wind speed: “Neither can be directly measured”, as he put it in an April 2023 post on the network’s real hashrate.
Work divided by time: what Bitcoin Core computes
Bitcoin Core ships its own estimator, the getnetworkhashps command. Its help text says it returns “the estimated network hashes per second based on the last n blocks”, with 120 blocks by default, or every block since the last difficulty change if you pass -1. The code in src/rpc/mining.cpp walks back over those blocks, takes the earliest and latest timestamps it finds among them, and divides the chain work accumulated over the window by that time span. Earliest and latest, not first and last: miners set block timestamps themselves, within limits, so timestamps are not always in order.
Every public tracker does a version of the same thing. The differences lie in the window, in how timestamps are treated and in whether the result is smoothed again before it reaches a chart. None of them adds information the chain does not contain, which is why none of them can be more “correct” than the others on a given day.
Short windows swing wildly, long windows lag
Block discovery is random. On average a block comes every ten minutes, but a gap of one minute or of forty minutes is perfectly normal, and the estimator cannot tell a lucky streak from a surge in machines. Lopp ran getnetworkhashps at block 784,978, in April 2023, when the real hashrate was around 350 EH/s. Over the last single block, the estimate came out at 4,470 EH/s. Over five blocks it was 375 EH/s, over ten 334, over 1,000 blocks 342. Over 10,000 blocks, about ten weeks, it dropped to 322 EH/s, because that window still remembered a smaller network.
That is the trade-off every chart makes. A one-block estimate can be off by a factor of 500, Lopp found, and a ten-block one is generally within a factor of three, which is still useless. A day of blocks, about 144, gives figures that move by tens of percent from one day to the next, as the Christmas 2022 series shows. A month of blocks is steady but tells you late that something has changed.
Why mempool.space, Hashrate Index and others disagree
Two trackers looking at the same chain at the same moment will show different numbers because they average over different windows. Lopp lists the windows of Hashrate Index: 432 blocks (three days), 1,008 blocks (seven days, which he calls the industry standard) and 4,320 blocks (thirty days). Kraken built a “True Hashrate” that wraps a 30-day rolling average in a 95% confidence band; for the period Lopp pictured, the band was nearly 40% as wide as the daily estimate itself. Pools add yet another figure, the sum of what their dashboards claim, which at block 785,000 came to 362 EH/s.
| Window | Blocks (about) | What it is good for | Weakness |
|---|---|---|---|
| 1 block | 1 | nothing | can be off by 500 times (Lopp, April 2023) |
| 120 blocks (Bitcoin Core default) | 120 | a quick look from your own node | swings of tens of percent |
| 1 day | 144 | spotting a sudden shutdown | noisy: 156 then 299 EH/s within four days in December 2022 |
| 3 days (mempool.space headline) | 432 | today’s level | still moves several percent a day |
| 7 days | 1,008 | the usual industry reference | reacts to a shock only after a few days |
| 30 days | 4,320 | trends and hash ribbons | lags real changes by weeks |
The 8 November 2026 figures show the effect in miniature. mempool.space’s daily points for the week ran 1,128 EH/s on 2 November, 910 on 3 November, then 1,034, 1,077, 955 and 1,001, with 989 for the partial day of 8 November. Averaged over the seven days, that is about 1,013 EH/s (our arithmetic). Nothing dramatic happened on 3 November; the blocks were simply slower that day.
When miners switch off, blocks slow down first
Difficulty only changes at the end of a 2,016-block period. Between two adjustments, the bar stays where it was set, whatever happens to the machines. If a quarter of the hashrate disappears, the remaining three quarters still have to clear a bar calibrated for the whole fleet, and blocks arrive more slowly until the next adjustment catches up.
The expected block interval follows directly: 600 seconds multiplied by the hashrate the difficulty was set for, divided by the hashrate actually working. Our arithmetic gives the scale of it. A 28% drop stretches the average block to about 13.9 minutes. A 40% drop gives 16.7 minutes, a 50% drop 20 minutes. Worse, the next adjustment then arrives later too: 2,016 blocks at 20 minutes each take 28 days, not 14.
It has happened. In the period that ended on 3 July 2021, while China’s miners were being shut down, blocks averaged 13.88 minutes, according to the timestamps of the two adjustments in mempool.space’s records (our arithmetic). The period took about 19.4 days instead of 14.
For someone sending bitcoin, a hashrate drop means waiting, not losing money. Fewer blocks per hour means less block space per hour, so the queue of unconfirmed transactions grows and fees tend to rise, the dynamics explained in our lesson on the mempool and transaction fees. Six confirmations that usually take about an hour can take an hour and a half or more. Confirmed transactions stay confirmed, and coins do not move without a valid signature, whatever the hashrate does. The security question, how much work an attacker would need to rewrite blocks, is about the total work behind the chain, which a few slow weeks hardly dent.
From 1 EH/s to a zettahash: ten years in one table
The genesis block of 3 January 2009 had difficulty 1, which any laptop could meet. The table below takes mempool.space’s daily hashrate series and its list of difficulty adjustments, both read on 8 November 2026. Daily figures are noisy, so treat each as a mark on a ruler, not a measurement. The difficulty column gives the value in force on that day, with the date it was set.
| Date | Daily hashrate estimate | Difficulty in force (set on) | What happened |
|---|---|---|---|
| 24 January 2016 | 1.02 EH/s | 0.11 T (13 January 2016) | first day above 1 EH/s |
| 4 October 2017 | 10.3 EH/s | 1.12 T (2 October 2017) | first day above 10 EH/s |
| 19 September 2019 | 104 EH/s | 11.89 T (13 September 2019) | first day above 100 EH/s |
| 10 May 2021 | 190 EH/s | 20.61 T (1 May 2021) | peak before China’s crackdown |
| 5 July 2021 | 83 EH/s | 14.36 T (3 July 2021) | after the largest cut ever, -27.94% |
| 25 December 2022 | 156 EH/s | 35.36 T (19 December 2022) | US winter storm |
| 16 September 2023 | 501 EH/s | 54.15 T (6 September 2023) | first day above 500 EH/s |
| 20 April 2024 | 642 EH/s | 86.39 T (10 April 2024) | fourth halving |
| 5 April 2025 | 1,025 EH/s | 113.76 T (23 March 2025), raised to 121.51 T that afternoon | first day above 1,000 EH/s in this series |
| 29 October 2025 | about 1,150 EH/s (7-day peak, 19 October) | 155.97 T (29 October 2025) | record difficulty |
| 25 January 2026 | 706 EH/s | 141.67 T (22 January 2026) | US cold wave |
| 8 November 2026 | about 1,000 EH/s (3-day) | 148.36 T (29 October 2026) | today |
Two milestones need a word of care. The zettahash, 1,000 EH/s, was first touched on Friday 4 April 2025 according to Glassnode data reported by CoinDesk, and on 5 April in mempool.space’s daily series. It slipped back below the following days. The level only held on a seven-day average from late summer: mempool.space’s daily points give a first seven-day mean above 1,000 EH/s on 31 August 2025 (our arithmetic), while Glassnode’s seven-day average first reached it in data CoinDesk reported on 2 September 2025. The single highest day in mempool.space’s series, 1,306 EH/s on 25 October 2025, is best read as a noisy day inside a seven-day peak of about 1,153 EH/s on 19 October.
The difficulty record, 155.97 T at block 921,312 on 29 October 2025, still stood on 8 November 2026. Today’s 148.36 T is 4.9% below it. The flat year in between included the biggest cut since 2021, -11.16% on 7 February 2026, and the biggest rise since May 2021, +14.73% twelve days later, as the machines switched off in January came back.
Three shocks the difficulty had to absorb
Most of the time hashrate drifts, and the adjustment follows in steps of a few percent. Three episodes show what happens when it does not drift but drops.
China’s crackdown, May to July 2021
On 21 May 2021, China’s State Council Financial Stability and Development Committee, chaired by Vice Premier Liu He, called for a crackdown on bitcoin mining and trading, according to the official readout of its 51st meeting on the Chinese government’s website. China had been the centre of mining, and its share was already falling: the Cambridge Centre for Alternative Finance, using geolocation data from four large pools, put its share at 75.5% in September 2019 and 46% in April 2021, before the restrictions began. Within days, BTC.TOP announced it was suspending its China business and HashCow said it would stop buying new rigs, Al Jazeera reported on 24 May.
mempool.space’s daily series falls from 190 EH/s on 10 May 2021 to 83 EH/s on 5 July, a drop of about 56% on those two days, though other trackers with other windows give other figures. Difficulty followed in three steps: -15.97% on 30 May, -5.30% on 13 June and -27.94% on 3 July at block 689,472, the largest cut in Bitcoin’s history, then -4.81% on 17 July. The machines were not destroyed; they were shipped. Cambridge’s update of 13 October 2021, with data to the end of August, found that declared mining in mainland China had effectively dropped to zero, and put the US first at 35.4%, ahead of Kazakhstan (18.1%) and Russia (11%).
The Christmas storm of 2022
In the last days of December 2022, an Arctic storm pushed power demand up across North America; Texas’s grid operator later reviewed it as Winter Storm Elliott in its public report on the event. Lee Bratcher, president of the Texas Blockchain Council, wrote on LinkedIn that 99% of industrial-scale bitcoin mining load in Texas was turned off at 6 a.m. on Saturday 24 December, CoinDesk reported, adding that Riot and Core Scientific curtailed operations.
The daily estimate went from 245 EH/s on 21 December to 156 EH/s on 25 December and back to 247 EH/s the next day. The period was short, so difficulty barely noticed: the adjustment of 3 January 2023 was -3.59%, after blocks averaged 10.39 minutes over the whole period (our arithmetic). A shock of a few days is absorbed by the average. That is the useful lesson from Elliott: what counts for the adjustment is how long machines stay off, not how far a daily chart falls.
The January 2026 cold wave
The same pattern returned on a larger network. Over the weekend of 24 and 25 January 2026, a winter storm across some three dozen US states took miners offline. Cointelegraph reported, from CoinWarz data, a seven-month low of 663 EH/s on Sunday 25 January, a fall of more than 40% in two days, with about 854 EH/s by Monday. US miners then ran nearly 38% of global hashrate, per Hashrate Index. CryptoQuant figures cited in the same report showed MARA’s daily production falling from 45 BTC to 7 BTC.
mempool.space’s series shows 1,159 EH/s on 23 January and 706 EH/s on 25 January. This time the outage overlapped a fresh period that had begun on 22 January, and machines returned slowly. Blocks in that period averaged 11.26 minutes, and the adjustment of 7 February 2026 cut difficulty by 11.16%, the largest cut since July 2021. When the fleet came back, the next one, on 19 February, raised it by 14.73%.
Hash ribbons: a named gauge of miner capitulation
Analysts use the hashrate series as a reading of miners’ financial stress. The best-known tool is the Hash Ribbons indicator devised by Charles Edwards, founder of Capriole Investments. It compares two simple moving averages of hashrate, 30 days and 60 days. When the 30-day average falls below the 60-day one, the indicator marks a “capitulation”: miners are switching off machines, usually because revenue no longer covers costs. When it crosses back above, the capitulation is considered over. Edwards describes capitulation as a measured decline in hashrate of roughly 10% to 40%, lasting one week to two months, according to Bitcoin Magazine in July 2022.
Edwards also presents the end of a capitulation as a buy signal, and his firm promotes its track record. Those are the author’s claims, and Bitcoin Magazine, in the same article, worked through the signal of 7 August 2021: a purchase made then had lost more than half its value by July 2022. Coindaro does not use the indicator as advice. As a description of the network, though, it is useful, provided you remember it lags: by construction it confirms that miners have already switched off.
It also depends on the input. Computed on mempool.space’s daily series (our arithmetic, plain 30- and 60-day means), the 30-day average was below the 60-day one from 6 January to 27 February 2026, covering the cold wave and the February cut, and has been above it since 27 August 2026; on 8 November the two stood at about 1,040 and 1,000 EH/s. Other implementations, with other hashrate series or extra price filters, give other dates.
Why miners switch off is mostly a matter of revenue per unit of hashrate, explained in the mining guide’s section on hashprice. Hashrate tends to follow miners’ margins, which follow the bitcoin price with a delay. The reverse link, hashrate driving price, has no established mechanism, and charts that overlay the two show correlation, not cause.
How other chains adjust difficulty
Bitcoin’s two-week step works because its hashrate is huge and fairly steady. Smaller proof-of-work chains, and chains that share an algorithm with a bigger one, cannot wait two weeks: if miners leave, blocks could stop for days. Each built its own controller, and the differences are a good map of the trade-offs.
Bitcoin Cash: from an emergency brake to ASERT
Bitcoin Cash split from Bitcoin on 1 August 2017, a story told in our lesson on forks and chain splits. It started with a fraction of Bitcoin’s miners and the same SHA-256 algorithm, so it added an Emergency Difficulty Adjustment. Bitcoin ABC’s source code describes it plainly: if producing the last six blocks took more than 12 hours, raise the target by a quarter, which cuts difficulty by 20%. The stated aim was to keep the chain from getting stuck “in case we lose hashrate abruptly”.
It worked too well. Because cheaper difficulty attracted miners from Bitcoin, who left again once it rose, hashrate swung between the two chains and BCH blocks came in bursts. A replacement went live on 13 November 2017: a per-block algorithm over a window of about 144 blocks, with the timespan clamped between 12 and 48 hours, per the upgrade specification. A simple moving average of that kind still oscillated, so on 15 November 2020 Bitcoin Cash moved to ASERT (aserti3-2d), written by freetrader, Jonathan Toomim, Calin Culianu and Mark Lundeberg. Its specification sets a target of 600 seconds and a half-life of 172,800 seconds, two days: each block’s target depends on how far the chain is ahead of or behind schedule since a fixed anchor block, with every two days of deviation doubling or halving it. The spec’s goal was to “eliminate periodic oscillations in difficulty and hashrate”.
Dogecoin: DigiShield retargets every block
Dogecoin started with a four-hour retarget for one-minute blocks. From block 145,000, its source code switches to DigiShield, credited in a comment to RealSolid and WDC: the difficulty is recalculated at every block, but only one eighth of the measured deviation is applied, a damping the code calls an amplitude filter, and each step is clamped between 0.75 and 1.5 times the target timespan. Since block 371,337, in September 2014, Dogecoin has also been merge-mined with Litecoin, which brought it far more hashrate than it could attract alone.
Litecoin: Bitcoin’s rule with the off-by-one fixed
Litecoin kept Bitcoin’s design: 2,016 blocks per period, which at 2.5-minute blocks is 3.5 days, and the factor-of-four clamp. One line differs. Where Bitcoin measures a period over 2,015 block intervals, Litecoin’s pow.cpp goes back the full period, a fix its comment credits to Art Forz and justifies as closing “an issue where a 51% attack can change difficulty at will”. It is the timewarp problem described in our proof-of-work lesson, solved in Litecoin’s code years before Bitcoin’s own proposal, BIP 54.
Monero: a trimmed 720-block window
Monero recalculates at every block too, with a statistical twist. Its difficulty code takes a window of 720 blocks, about a day at the two-minute target, leaving out the 15 most recent. It sorts their timestamps, drops the 60 lowest and 60 highest, and sets the next difficulty so that the work done over the remaining span would have taken exactly two minutes per block. Throwing away outliers at both ends makes the result hard for any miner to push with fake timestamps. The price is speed: a window a day long, starting 15 blocks back, takes time to register a sudden change.
Ethereum before the Merge: a bomb delayed six times
Proof-of-work Ethereum adjusted difficulty at every block from the gap since its parent. Under the Homestead rule of EIP-2, each block moved difficulty by a 2,048th step per ten seconds of deviation, and from Byzantium in October 2017 the formula of EIP-100 also counted uncle blocks. On top of that sat a deliberate time bomb: a term that doubled every 100,000 blocks, designed to make mining slower and slower and so push the community towards proof of stake.
Each time it began to bite, or was about to, an upgrade pushed it back. Six upgrades did so, each with its own EIP, dated on ethereum.org’s history page: Byzantium (EIP-649, 2017), Constantinople (EIP-1234, 2019), Muir Glacier (EIP-2384, 2020), London (EIP-3554, 2021), Arrow Glacier (EIP-4345, 2021) and Gray Glacier (EIP-5133, 30 June 2022). Gray Glacier’s EIP-5133 delayed it by a further 700,000 blocks “to avoid network degradation due to a premature activation of the difficulty bomb”, to mid-September 2022. The Merge arrived first: Ethereum switched to proof of stake on 15 September 2022: the first proof-of-stake block, 15,537,394, came at 06:42:59 UTC, after total difficulty had passed 58,750,000,000,000,000,000,000. Ethereum has had no mining difficulty since.
| Chain | How difficulty adjusts | Window | Target block time | Since |
|---|---|---|---|---|
| Bitcoin | fixed periods, factor-of-4 clamp | 2,016 blocks | 10 minutes | 2009 |
| Bitcoin Cash | ASERT aserti3-2d, per block | exponential, 2-day half-life | 10 minutes | 15 November 2020 |
| Litecoin | fixed periods, full-period lookback | 2,016 blocks (3.5 days) | 2.5 minutes | 2011 |
| Dogecoin | DigiShield, per block, 1/8 damping | 1 block | 1 minute | block 145,000 |
| Monero | per block, trimmed window | 720 blocks minus 60 at each end | 2 minutes | current code |
| Ethereum (to 2022) | per block plus difficulty bomb | 1 block | no fixed target: gaps of 10 to 19 seconds left difficulty unchanged | ended 15 September 2022 |
Bitcoin’s hashrate and difficulty on 8 November 2026
The figures below come from mempool.space’s public API, read on 8 November 2026 at about 09:59 UTC. Projections move with every block: the adjustment of 13 November will probably have happened by the time some readers see this page, and our guide to reading a block explorer shows where to check the live values yourself.
| Figure (8 November 2026, 09:59 UTC) | Value |
|---|---|
| Latest block | 975,047 |
| Difficulty | 148,364,561,003,867 (148.36 T), set at block 973,728 on 29 October 2026, +5.54% |
| Estimated hashrate, 3 days (mempool.space) | about 1,000 EH/s |
| Daily estimates, 2 to 8 November | 1,128; 910; 1,034; 1,077; 955; 1,001; 989 EH/s (last day partial) |
| 7-day mean of those daily estimates (our arithmetic) | about 1,013 EH/s |
| 30-day and 60-day means (our arithmetic) | about 1,040 and 1,000 EH/s |
| Average block time in the current period | about 10 minutes 30 seconds |
| Progress of the period | 65.4%, 697 blocks to go |
| Next adjustment | block 975,744, estimated 13 November 2026 around 12:00 UTC, projected -4.7% |
| Record difficulty | 155.97 T (29 October 2025), 4.9% above today |
Read together, the lines say one thing. Difficulty was set on 29 October for a network doing about 1,062 EH/s; the network has delivered about 1,000 since; blocks have run about 5% slow; and the projected cut of about 4.7% is the adjustment closing that gap. Over the seven days to 8 November, Foundry USA found 226 of 936 blocks, about 24%, per mempool.space: the concentration of pools is a separate question, covered in the mining guide.
What US-listed miners report to the SEC
For American readers, the most detailed hashrate figures that come with legal liability attached are those of listed mining companies, in their filings with the Securities and Exchange Commission. MARA Holdings’ annual report for 2025 (Form 10-K) gives an “energized hashrate” of about 66.4 EH/s at 31 December 2025, up from 53.2 EH/s a year earlier, of which 37.2 EH/s at owned sites and 29.2 EH/s at hosted ones. It says it mined 8,799 bitcoin in 2025, and that it has begun adding AI inference work at its Granbury site in Texas.
The definition matters. MARA defines energized hashrate as what its rigs could theoretically produce if all of them ran at 100% of the manufacturers’ specifications. It is a capacity figure in the company’s own terms, not a network measurement. Set against mempool.space’s daily estimate of about 979 EH/s for 1 January 2026, it would be about 7% of the network (our rough arithmetic, mixing two different kinds of number). The share of blocks a company actually wins, visible on chain through its pool, is the better check.
Six things to keep from this lesson
- Difficulty is exact and public; the bitcoin hashrate is an estimate derived from block times at that difficulty.
- The shorter the window, the noisier the estimate: one block can be off 500 times, one day by tens of percent, thirty days lags by weeks.
- A sudden loss of hashrate shows up first as slower blocks and higher fees, and in difficulty only at the next 2,016-block boundary.
- The largest cut on record, -27.94% on 3 July 2021, followed China’s crackdown; US winter storms in December 2022 and January 2026 left smaller marks.
- Hash ribbons compare 30- and 60-day averages of hashrate to flag miner capitulation; they confirm, they do not predict.
- Bitcoin Cash, Dogecoin and Monero adjust at every block, Litecoin fixed Bitcoin’s off-by-one, and Ethereum left mining for proof of stake in 2022.
The next lesson in this path turns to proof of stake, where security comes from locked coins rather than work, and there is no difficulty to adjust. Bitcoin’s next adjustment, at block 975,744, is due around 13 November 2026.
