Bitcoin hash rate: what it measures, why it is only an estimate, and why it matters

Hash rate counts guesses per second, across every miner at once

A hash is the output of a function that scrambles data into a fixed-length number. Bitcoin mining means running a block header through that function, SHA-256 applied twice, over and over with a small change each time, hoping to land on a result at or below the network’s target. Each attempt is one hash. The Bitcoin Wiki defines hash per second as the number of double SHA-256 computations performed in one second, and calls the total across all miners the hash rate.

The numbers are enormous, so they get prefixes. The wiki spells out a petahash as a quadrillion hashes; an exahash is a thousand petahashes, and a zettahash is a thousand exahashes, which CoinDesk states as one zettahash equaling 1,000 exahashes. Most charts today quote the network in exahashes per second, written EH/s. The prefixes are the same ones used for computer storage, and they climb the same way: by a factor of a thousand each step.

What hash rate does not count is people, machines or places. It is a single global number with no location attached. A rise could be new machines in one warehouse or a thousand small ones; the number cannot tell you which. Our guide to how mining works covers the hardware, and the mining pools article covers how individual miners combine their guesses.

Nobody measures it; everybody infers it

There is no meter on the network. Miners do not report their attempts, and there is no central place to report them to. What the chain records is the difficulty in force and the timestamp of each block. From those two, anyone can work backward: if the target was this hard and blocks are arriving this fast, the network must be guessing at roughly this rate. Bitcoin Core’s own command for it, getnetworkhashps, is described in its documentation as returning the estimated network hashes per second based on the last n blocks, 120 by default.

Data providers do the same thing over different windows. Coin Metrics states that its hash rate figure is derived from the mean difficulty and the mean block interval and gives an estimate of how much hash power is mining a chain. The Cambridge Centre for Alternative Finance, whose electricity index depends on hash rate, takes its daily figure from Coin Metrics rather than measuring anything itself, and notes that actual power demand cannot be determined because of the decentralized nature of the network.

Because block discovery is random, the estimate is noisy. A lucky day with more blocks than usual looks like a surge in hash rate even if nothing changed. Blockchain.com’s chart notes carry exactly that warning, that daily raw values may rise or drop as a result of the randomness of block discovery, and recommend a seven-day average for a truer picture. This is why two websites can show different hash rates at the same moment and both be defensible. Different averaging windows, timestamp treatment and data inputs can produce different estimates.

The math fits in one line

The Bitcoin Wiki gives the shortcut: hash rate is approximately the difficulty multiplied by 2^32 and divided by 600. The 2^32 comes from the structure of the target; at difficulty 1, a miner needs about that many guesses on average to find a valid block. The 600 is the number of seconds in the ten-minute target between blocks. So difficulty tells you how many guesses a block should take, and dividing by the intended time between blocks turns that into guesses per second. At difficulty 1, the wiki notes, that comes to around seven million hashes per second.

That formula assumes blocks are arriving exactly on schedule. Real estimates replace the 600 with the average interval actually observed over the chosen window, which is where the differences between providers come from. It also means hash rate and difficulty are two views of the same thing. Difficulty is a consensus parameter adjusted from elapsed block time every 2,016 blocks, not a direct measurement of mining hardware. Hash-rate charts estimate activity from work and observed timing.

From petahashes to zettahashes: the growth, with dates

The scale has changed by a factor that is hard to hold in your head. The Bitcoin Wiki records the network at around 300 petahashes per second in January 2015, which it spells out as 300 quadrillion hashes per second, and at around 170,000 petahashes per second in April 2021. CoinDesk, citing Glassnode data, reported the seven-day average crossing one zettahash per second for the first time on 2 September 2025, having climbed from around 800 exahashes per second at the start of that year. One zettahash is a million petahashes, so the network grew more than a thousandfold between January 2015 and September 2025.

We deliberately give no current figure. Hash rate moves every day, and any number written here would be stale before long. A chart from a provider that explains its method and shows its averaging window will tell you where things stand today. What the dated figures above show is the direction and the scale.

Why hash rate is a security number

The white paper explains the idea in a few sentences. Proof-of-work is essentially one CPU, one vote, and if a majority of CPU power is controlled by honest nodes, the honest chain will grow the fastest and outpace any competing chains. To modify a past block, an attacker would have to redo the proof-of-work of that block and all the blocks after it, then catch up with and surpass the honest network. The more guessing power the honest side has, the more an attacker needs, and the more it costs to acquire.

Hash rate is therefore a rough price tag on rewriting history. It is not a guarantee. What matters is how the power is distributed, and a number that is high but concentrated in a few hands is less reassuring than the total suggests. Our article on the 51% attack works through what a majority could and could not do. It is also worth knowing that hash rate is not comparable between networks with different hash functions, a point Coin Metrics makes in its documentation.

Finally, guesses cost electricity, which is where hash rate meets the energy question. Cambridge’s index converts hash rate into an estimated power draw by assuming a mix of mining hardware and its efficiency, and it publishes a floor and a ceiling rather than one number precisely because it, too, is estimating from the outside. Every figure in this subject, from the hashes to the watts, is inferred. That does not make them wrong; it makes them estimates, and they should be read as such.