How does Bitcoin mining work? The mechanism, step by step

Bitcoin is a public ledger of every transaction, copied across thousands of computers, the nodes A computer that keeps a full copy of Bitcoin and checks every block itself. Run your own node. The problem is easy to state: who gets to write the next page, with no bank or authority to decide, and without a cheat being able to rewrite the past to spend the same bitcoin twice?

The answer in Satoshi Nakamoto’s white paper is called proof of work: to propose a block, you must show you spent a lot of computation finding it. Rewriting an old block would mean redoing all that work, then the work of every block after it, faster than the rest of the network. The white paper sums it up: proof of work is “one-CPU-one-vote”.

So mining does two jobs at once: it secures the ledger and it issues new bitcoin, paid to whoever finds the block. If you are just starting, the guide to getting started with Bitcoin mining gives the short version; here, we open the hood.

When you send bitcoin, your transaction is broadcast to the network’s nodes, which check it and keep it in their mempool The waiting room for transactions not yet written into a block. Whoever builds the block chooses which ones go in., the waiting room for transactions not yet written into a block. On 28 October 2026 around 11:00 UTC, 47,784 transactions were waiting there according to mempool.space, about 20 million vbytes The unit of size of a transaction. Fees are paid per vbyte, not according to the amount sent.: enough to fill some twenty blocks, each holding at most one million.

A block has a size limit (a weight of 4 million units in Bitcoin Core’s code), so choices must be made. The miner’s node builds a block template, taking first the transactions that pay the highest fee per vbyte. At the top it adds a special transaction, the coinbase The first transaction of a block, the one that pays the reward. On a solo pool it pays your address directly., which pays the reward to the miner: it is the only way new bitcoin come into existence.

A real example: block 973,541, mined on 28 October 2026, holds 6,061 transactions and weighs 3,997,625 units out of the 4 million allowed. In a pool A server that machines connect to. A regular pool shares the earnings, a solo pool does not. Solo or pool?, the pool builds this template from its node and sends it to the machines; when you mine with your own node, you choose the transactions yourself (solo mining with Bitcoin Core).

Your miner never sees the thousands of transactions. It only works on the block’s header: 80 bytes, six fields, always in the same order. The transactions are summed up in a single hash A 64-character code computed from the block. Impossible to predict: you have to try., the Merkle root: changing one transaction would change that root, so the header, so the whole calculation.

The six fields of a Bitcoin block header, with the real values of block 973,541 (28 October 2026, mempool.space). Sizes and roles: bitcoin.org developer reference.

Field	Size	What it is for	In block 973,541
Version	4 bytes	The validation rules followed. Some of its bits also serve as spare search space (“version rolling”).	0x3a222000
Previous block hash	32 bytes	The link to the block before: changing an old block would break everything after it.	00000000000000000000a186…dcc58c (block 973,540)
Merkle root	32 bytes	A single hash that sums up every transaction in the block.	1f7617e1…92532c, for 6,061 transactions
Time	4 bytes	The time stated by the miner, later than the median of the previous 11 blocks.	28 October 2026, 11:07 UTC
nBits	4 bytes	The target, in a compact form: the block’s hash must be less than or equal to it.	0x17020093
Nonce The number the machine changes on every attempt to get a new hash.	4 bytes	The number the machine changes on every attempt.	1,613,434,802
Total	80 bytes	This small piece of data, and only this, is what the machine runs twice through SHA-256 The hash calculation Bitcoin uses, and so mining. Bitcoin Cash uses the same one: the same machines mine either..	hash: 00000000000000000001c527…69a3c

The “previous block hash” field is what makes a chain of blocks: each block contains the hash of the one before, which contains the one before that, back to the first block in 2009. That is why nobody can quietly change an old block.

For each attempt, the machine runs the header through the SHA-256 function, then runs the result through SHA-256 again: that is double SHA-256. Out comes a 256-bit hash, written as 64 hexadecimal characters, impossible to predict: flipping a single bit of the header gives a completely different hash. So you cannot aim for a good hash, only try again and again.

What changes from one attempt to the next is first the nonce, the header’s last field. It is 4 bytes: about 4.29 billion possible values. A lot for a human, very little for a chip: a Bitaxe Gamma at 1.2 TH/s (rated) makes about 1.2 trillion attempts per second, so it runs through every nonce in about 3.6 milliseconds, roughly 279 times a second.

Something else has to change then. The pool (or your own mining software) leaves a small free field in the coinbase transaction, the extra nonce: by changing it, the machine changes the coinbase, so the Merkle root, so the header, and starts again with 4.29 billion fresh nonces. Recent chips also roll a few bits of the version field, known as version rolling and provided for by BIP 320, and the timestamp can move forward a little.

You can check the principle yourself: the double SHA-256 of the 80 header bytes of block 973,541 gives exactly 00000000000000000001c527…69a3c, the hash shown by block explorers (which write it byte-reversed, by convention). And any node redoes that calculation in a split second: finding is very slow, checking is instant.

A hash is also a gigantic number. The network sets a limit, the target (encoded in the nBits field): a block is only valid if its hash is less than or equal to that target. Since the target is tiny compared with all possible hashes, a winning hash necessarily starts with a long run of zeros.

At the current difficulty (140.6 T, network as of 17 October 2026), it takes about 79 leading zero bits, or 19 or 20 zeros in hexadecimal: the hash of block 973,541 has 19, followed by a 1. Every attempt has the same small chance, and none brings the next one closer.

Every 2016 blocks, the network adjusts the target: Bitcoin Core’s code aims for 10 minutes per block and two weeks per period (14 days of 1,440 minutes, divided by 10 minutes, is 2016 blocks). If the last 2016 blocks came too fast, the target drops and difficulty rises; too slow, the opposite, with a change of at most a factor of 4 each time. That keeps the pace at about one block every ten minutes, however many machines are plugged in. The details, “shares” and your best share are in mining difficulty explained.

Whoever finds the block pays the reward to themselves in the coinbase transaction. It has two parts:

The subsidy started at 50 BTC in 2009 and is cut in half every 210,000 blocks, about four years: that is the halving, written into the code (Bitcoin Core’s GetBlockSubsidy function). The next one falls at block 1,050,000, five times 210,000; its date, around April 2028, is an estimate that depends on the actual pace of blocks.

These successive cuts cap the total at a little under 21 million bitcoin; on top of that, Bitcoin Core’s code rejects any amount above 21 million as a sanity check (MAX_MONEY). At block 973,541, about 895,000 were left to be issued, for every miner in the world. When the subsidy has become negligible, around 2140, only fees will remain, as the white paper already planned.

These bitcoin cannot be spent right away: they need 100 more blocks, about 16 hours 40 minutes. What actually happens when a small miner wins: found a block solo, what next?

With about 600,000 billion billion attempts needed per block on average, a single machine can wait for centuries, and a Bitaxe for millennia. Pools combine the hashrate of many machines: the pool asks them for hashes much easier than the target, called shares, which prove the work done, then splits the earnings according to that work. On average a machine earns the same in a pool as solo, minus the pool’s fee, but the earnings arrive more regularly: solo or pool, which one?

The five pools that found the most Bitcoin blocks over the twelve months to 28 October 2026: 52,369 blocks in all, attributed by mempool.space from the tags each block carries.

Pool	Blocks found in a year	Share of blocks
Foundry USA	14,679	28.0%
AntPool	9,572	18.3%
F2Pool	6,720	12.8%
ViaBTC	5,019	9.6%
SpiderPool	4,540	8.7%
The five together	40,530	77.4%

Two pools, Foundry USA and AntPool, found 46.3% of the year’s blocks, and the top five 77.4%. A pool does not own the machines mining on it, and every miner can switch by changing an address in their settings. But the pool builds the block template, so it chooses the transactions: this concentration is a real issue, which I covered in a video (in French), 6 companies control Bitcoin. Here is how.

Every machine mining solo, in someone’s home, on a pool that pays the reward straight to the miner’s own address, is a bit of hashrate outside these giants. That is also the idea behind FindMyBlock, the Bitcoin solo pool I built, with no fee: solo mining on FindMyBlock.

Because the energy spent is precisely what makes the ledger expensive to attack, and because difficulty adapts. As long as mining pays more than the electricity, new machines arrive, network hashrate rises, and the 2016-block adjustment raises difficulty to keep ten minutes per block. More efficient machines therefore do not necessarily lower total consumption: mostly, they raise the difficulty.

The network now runs at about 1,006 EH/s (network as of 17 October 2026), that is 1,006 billion billion hashes per second, a figure computed from the difficulty. Its consumption depends on the machines’ average efficiency, in joules per TH: at this hashrate, every J/TH of average efficiency is about 1.0 GW, around the clock. For example, if every machine on the network had the efficiency of a Thor P2 (15.3 J/TH, measured at the wall), the network would draw about 15 GW. The real fleet mixes recent and older machines: for a tracked, sourced estimate, the reference is the Cambridge Bitcoin Electricity Consumption Index (CBECI).

At home it is simpler: every watt drawn ends up as heat in the room. A 15 to 18 W Bitaxe Gamma costs at most about €32 of electricity a year at €0.20 per kWh, running day and night. The math for each machine: Bitcoin miner power consumption, and to put the heat to use: heat your home mining bitcoin.

A Bitaxe Gamma does exactly the same calculation as industrial farms, with the same chip as the Antminer S21 Pro, a recent industrial machine (the BM1370, according to the Bitaxe project repository): about 1.2 trillion double SHA-256 hashes per second (1.2 TH/s, rated). Against the network, that is about a 1 in 840 million chance at each block: it would take about 840 million Bitaxes to match Bitcoin’s whole hashrate.

Over a year, that comes to about a 1 in 16,000 chance of finding a block solo (network as of 17 October 2026). My shelf, 13 machines for about 50 TH/s, has about a 1 in 383 chance a year, and has not found a Bitcoin block yet. It is small, but not zero: every attempt has the same chance, whether the machine sits at home or in a farm, and small machines have already won whole blocks.

For the odds of your own machine, the solo mining calculator redoes this math. For how long it would take to build up one bitcoin, machine by machine: how long to mine one bitcoin. And to choose a machine: which Bitaxe to buy. To watch the mechanism run for a few euros, a NerdMiner does the same double SHA-256 on an ESP32 chip, millions of times slower: it is a learning tool, not a way to earn.

Network figures as of 17 October 2026, recalculated at each difficulty adjustment; example block, mempool and pool shares read on mempool.space on 28 October 2026. Technical words are in the glossary.