How Bitcoin mines a block

Auto-advance Speed 1× Show full nodes Read how it works

Time is compressed: a ten-minute block takes about eight seconds here at 1×. Click a step on the right, or use the arrow keys, to jump to it.

Entering the mempool

waiting in the mempool selected into a block template hashing, and a block just found verified, on the chain stale block, abandoned real transaction entering the mempool a real block, found by that pool on the chain, gaining confirmations inner ring: mining pools, sized by hashrate

A wallet hands a signed transaction to one node, which checks it and relays it to its peers. Within seconds nearly every node holds it in its mempool, the waiting room for unconfirmed transactions. Nothing is ordered yet.
Each miner assembles a candidate block from its own mempool, taking the transactions that pay the most per virtual byte first, until the block's limit of 4 million weight units (a little under 4 MB) is full. Transactions that pay too little wait, sometimes for days.
Every miner hashes its 80-byte block header with SHA-256, twice, changing the nonce each time and hoping for a result below the network's target. Trying nonce 00000000. The odds per hash are the same for everyone, so a miner with 25 % of the hashrate finds about 25 % of blocks: a lottery weighted by work.
One miner gets lucky and broadcasts the block. Each node that receives it checks the header and passes it on. Compact block relay sends only the header and short transaction IDs, since peers already hold most of the transactions, so the whole network hears within a few seconds.
Every node checks the proof of work, every signature and every input itself, trusting no one. A valid block joins the chain. When two miners find blocks seconds apart, the network briefly splits; the next block extends one tip and the other becomes stale.
Each block built on top is one more confirmation. Undoing a block means redoing its work and all the work after it, so the chance of a reversal fades fast. Six confirmations, about an hour, is the usual bar for a large payment.

Nodes don't publish the hashes miners try, so live mode shows blocks the moment mempool.space's node hears them, and transactions as they enter its mempool. Pick any step to study the race in step-by-step mode.

Who mined the last 40 blocks Loading recent blocks

How this is counted

Each column groups five consecutive blocks by the pool that mined them, read from the coinbase transaction using mempool.space's pool tags. Blocks whose coinbase matches no known pool count as Unknown. Forty blocks is under seven hours, so small pools appear and vanish from one window to the next.

Transactions per block Waiting for the first block

SegWit transactions
legacy (pre-SegWit) transactions

Block chain 0 tx in the mempool 0 EH/s 0 sat/vB for the next block 0 blocks to the retarget 0 blocks to the halving

The parts that make this work

Proof of work and the target

A block header holds the previous block's hash, a Merkle root committing to every transaction, a timestamp, the encoded target and a 32-bit nonce. SHA-256 applied twice to those 80 bytes gives a number that looks random; the block is valid only if that number is below the target. Nothing clever helps, only more attempts. Since the nonce alone runs out in seconds on modern hardware, miners also vary an extra nonce in the coinbase transaction and a few version bits. In September 2026 the network makes roughly a thousand quintillion attempts a second, about 1,000 EH/s.

Difficulty adjustment

Every 2,016 blocks each node compares how long those blocks actually took with the two weeks they should have taken and scales the target so the next 2,016 average ten minutes each. A single step can't move it more than four times either way. The current difficulty is about 133 trillion, which means a valid header is around 133 trillion times harder to find than it was for the first block in 2009. The count of blocks until the next retarget is on the strip above.

Why "longest" means most work

Nodes follow the chain whose blocks embody the most total work, summed from each block's target, not simply the one with the most blocks. That rule is what lets a node that has been offline, or is seeing two competing tips, decide on its own which history is the real one. Anyone wanting to replace a block has to out-work the entire honest network from that block onward, which is what proof of work is for.

Stale blocks, reorgs and six confirmations

Two miners sometimes find valid blocks within seconds of each other. Both spread, each node keeps the one it saw first, and the tie holds until the next block extends one of them. The other becomes stale: its transactions go back to the mempool and are almost always included in the next block. Reorganisations of one block happen a few times a year on mainnet; deeper ones are far rarer. Finality is probabilistic. With every block on top, reversing a payment gets harder, and six blocks, about an hour, is the customary point at which exchanges treat a large deposit as settled.

The block reward, halvings and the 21 million cap

The miner of a block writes a coinbase transaction paying itself the subsidy plus every fee in the block. The subsidy started at 50 BTC and halves every 210,000 blocks; it has been 3.125 BTC since block 840,000 in April 2024 and drops to 1.5625 BTC at block 1,050,000, expected in early-to-mid 2028. About 20.1 million of the 21 million bitcoin that will ever exist have been mined. Nothing is burned: fees go to miners, and the rest of the 21 million arrives ever more slowly until the subsidy rounds to zero around 2140.

Fees and the mempool market

Block space is fixed, so transactions compete for it on price per virtual byte. A node will not relay a transaction below its minimum fee rate (0.1 sat/vB on current Bitcoin Core by default), and miners take the best-paying transactions first. When the mempool is quiet the next block clears at about 1 sat/vB; during a rush the rate can climb into the hundreds. Wallets can replace an unconfirmed transaction with a higher-fee version (RBF), and a mempool that runs out of memory drops its cheapest transactions. Fees on a quiet day add well under a tenth of the subsidy to a block's reward.

Mining pools and where hashrate lives

An individual machine would wait years between blocks, so miners join pools: the pool assembles the block template, members hash it, and the reward is split by the shares of work each contributed. The pool's name is usually written into the coinbase, which is how the chart above tells them apart. Concentration is real: the two largest pools, Foundry USA and AntPool, found close to half of the blocks in the last week of September 2026. Pools coordinate hashrate but don't own it; members can point their machines elsewhere in minutes.

What live mode shows

Live mode connects to mempool.space's public WebSocket and REST API. Every violet dot is a real transaction entering that node's mempool, with its size and fee rate in the feed. When a block is found, the pool that mined it flashes in the inner ring, where each pool is sized by its share of the last week's blocks, and the block's height, hash, transaction count, fees and subsidy are real. The chart, the mempool count, the hashrate, the fee estimate, the retarget countdown and the distance to the halving all come from the same source. The outer ring of full nodes is symbolic: there are tens of thousands, and the propagation wave is drawn, not measured.

What the animation simplifies

Time is compressed about seventy-fold: a ten-minute block takes eight or nine seconds here. Real miners each build a slightly different template rather than sharing one, and they never stop hashing, even while a new block propagates. The step-by-step miners, their hashrate shares and the transaction counts are illustrative, and stale blocks are staged far more often than the real network produces them. Read the Bitcoin developer guide or the original whitepaper for the full protocol.