How Mining Works

The consensus mechanism, adjustment period, target interval, reward amount, and coinbase maturity are fixed protocol rules. Hash rate and electricity use are point-in-time estimates that move daily, so check the primary data in the sources below for current values.

A miner selects transactions it considers valid, constructs a candidate block, varies block-header data in search of proof of work, and proposes a qualifying candidate to the network. The coinbase of an accepted block assigns the newly issued bitcoin and transaction fees.

Miners do not decide validity on their own. Every full node independently checks the transactions, block structure, issuance, scripts, and proof of work against the consensus rules, and rejects invalid candidates. Mining attaches cost to proposed history in an open-participation network and supports issuance and ordering by cumulative work.

The name "mining" comes from gold mining. Like gold, Bitcoin must be "mined" by putting in resources: computing power and electricity.

Proof of Work (PoW) searches for a SHA-256d hash of a block header that is numerically below the network target. “Leading zeros” is a visual shorthand; the rule compares a 256-bit integer with the target. A valid hash is probabilistic evidence of the expected search work at the current target.

Miners vary more than the nonce: coinbase extra nonce data, the transaction set, timestamps, and other permitted header inputs create further candidates. The number of trials is not fixed; it is a random variable governed by success probability and network difficulty.

A miner that finds a below-target hash broadcasts a candidate block. Other nodes cheaply recompute SHA-256d and the target comparison, then separately validate transactions, scripts, issuance, the Merkle root, and every other consensus rule. A candidate can become stale in propagation and later history can reorganize, so finding a hash is not the same as final settlement.

PoW is "hard to compute, easy to verify" and that asymmetry is the source of the network's security.

Bitcoin is designed to produce one block roughly every 10 minutes. To hold that interval, the mining difficulty adjusts automatically every 2,016 blocks (about two weeks).

If blocks arrive too fast, difficulty rises; if too slowly, it falls. The correction is retrospective, however: the network looks at how long the previous 2,016 blocks actually took and only then sets the difficulty for the next 2,016. The rate is not "held constant" so much as allowed to drift and then pulled back toward 10 minutes. Each adjustment is clamped to a 1/4x–4x range, so difficulty moves only in steps even when hash rate swings sharply.

Because the correction lags, the measured average interval is not exactly 10 minutes. From the genesis block through block 962,733 the measured average is about 9.62 minutes, or 17 years of running slightly fast.

Difficulty is expressed as a "target hash value." The hash a miner finds must be smaller than that target, and the smaller the target, the higher the difficulty.

As of 2026, difficulty stands at roughly 130 trillion times its initial 2009 value, a measure of how far mining power has grown in 17 years (difficulty adjusts up or down every 2,016 blocks and has recently hovered around the 100-trillion-times range).

Two miners sometimes find a valid block at almost the same moment. The network splits briefly into two branches, and as soon as the next block lands on one of them, the shorter branch is discarded. The discarded block was neither invalid nor fraudulent; it simply lost the race. Blocks like this are known as stale, or orphan, blocks, and the miner who produced one loses the reward.

The chance of producing a stale block depends on how quickly a block propagates across the network, which is why well-connected, larger miners are often said to have an edge. A strategy that deliberately exploits this asymmetry, withholding a found block for a while so that rival miners waste work, was described theoretically in 2013 as "selfish mining," though no clear instance of it being executed on Bitcoin mainnet has been documented.

This is also why confirmations matter. The most recent blocks are the ones most likely to be replaced, so larger payments are worth more confirmations. For the same reason, a mined block reward (the coinbase) cannot be spent until 100 blocks have passed, a rule that guards against exactly this kind of chain reorganization.

Early Bitcoin clients contained a CPU miner. In 2010, contributors shared an SSE2 implementation that parallelized SHA-256 on Intel and AMD CPUs, followed that October by a public OpenCL miner used with ATI and NVIDIA GPUs. Open-source designs targeted FPGAs from the then-independent Xilinx and Altera in 2011, while Canaan’s SEC filing records shipment of Avalon ASIC machines in January 2013. These transitions overlapped; participants did not all switch device classes on one date.

A CPU executes broad software, a GPU parallelizes many trials, an FPGA maps the SHA-256 data path into reconfigurable logic, and an ASIC fixes it in silicon. A stable algorithm and revenue tied to hash rate against electricity and capital costs rewarded specialization. Claims that a GPU or ASIC is simply some fixed number of times “faster” hide the application, the precision, and the measurement boundary.

Competitive Bitcoin mining now centers on SHA-256 ASIC systems, but a system is more than an ASIC die: it needs controllers, boards, memory, power supplies, firmware, air or liquid cooling, networking, and a facility. System firms including Canaan, Bitmain, and MicroBT, foundries, packaging and test suppliers, pools, and operators occupy distinct roles. Intel also entered with Blockscale in 2022 but issued a product-discontinuance notice in 2023; entry by a large company and business persistence are different facts. Chip J/TH, wall-level miner efficiency, and facility efficiency are different measurements.

The full lineage, from CPU, GPU, FPGA, and ASIC through programmability, parallelism, data movement, manufacturing, and corporate roles, is covered in “Computing Hardware.”

A mining pool is an arrangement in which many miners combine their computing power to raise the odds of finding blocks. Rewards are then split among the participants in proportion to the power each contributed.

For a miner working alone, the chances resemble a lottery. In a pool the payouts are small but steady.

Measured by blocks mined over the trailing month (as of August 2026), the ranking is Foundry USA 1,132, AntPool 854, F2Pool 741, SpiderPool 377, ViaBTC 361, MARA Pool 199, SECPOOL 187, and Luxor 144, with the top two pools alone accounting for roughly 45% of all blocks. Rankings and shares shift quickly as operators exit or merge and miners move, so treat these as point-in-time figures.

Pool concentration is a concern for network decentralization. If a single pool exceeds 50% of the hash rate, a 51% attack becomes theoretically possible. Participants can switch pools at any time, which is a real check, but it does not resolve the problem. Under Stratum V1, the widely used pool protocol, the pool rather than the individual miner decides what goes into a block (the block template); miners merely lend hash power. Censoring specific transactions therefore works for as long as it takes miners to notice and move. Stratum V2 is the proposal that hands template construction back to miners, and its adoption is still in progress.

Bitcoin mining's annual electricity consumption is comparable to that of some countries. On the standard reference index, the Cambridge Bitcoin Electricity Consumption Index (CBECI), the best-guess estimate is approximately 141 TWh per year as of early August 2026.

That figure is an estimate, not a measurement. CBECI models the mix of hardware miners are running, so pushing that assumption toward "everyone runs the most efficient machines" or "everyone runs the least efficient" produces a lower and upper bound spanning roughly 75–250 TWh. Numbers of this kind should always be read together with their band and their as-of date rather than as a single settled value.

Critics call it "a waste of energy," but proponents counter that it is "the necessary cost of maintaining value security."

Renewable energy use has been rising. Regions with abundant hydropower (Iceland, Canada, Norway) and flare gas mining that burns off excess natural gas have both drawn attention.

The sustainable-energy share depends heavily on who measured it and how. The Bitcoin Mining Council (BMC), an industry body, reports approximately 60%, but that figure aggregates self-reported data from member firms and covers only about half the network. Cambridge's independent estimate puts the share at roughly 52.4%. The two differ in what counts as "sustainable" (whether nuclear is included, for instance) and in how much of the network they cover, so they should be compared with their measurement conditions attached rather than quoted one at a time. Improving that mix remains a priority for the industry.

Miner revenue comes from two sources: ① the block reward (newly issued BTC) and ② transaction fees (the sum of the fees on the transactions in the block).

Block rewards halve at each halving. After the fourth halving in 2024, the reward is 3.125 BTC. Once all bitcoin has been issued, around 2140, fees alone will make up miner revenue.

Transaction fees rise and fall with network congestion. During peak periods in 2017 and 2021, fees sometimes exceeded $50 per transaction; the highest on record is an average of about $128 per transaction on April 20, 2024, when the fourth halving coincided with the launch of Runes (an Ordinals-derived token standard).

Whether fee revenue alone can sustain miner incentives over the long run is one of the open questions about Bitcoin's sustainability.

For SHA-256 search, Grover's algorithm offers an idealized square-root query speed-up. That is not the same as a quantum miner instantly gaining majority hash power. Fault-tolerant circuits, error correction, execution rate, power, parallelism, and competition with classical ASICs are separate engineering constraints, and no practical quantum mining attack has been publicly demonstrated.

Signatures are a different surface. If a sufficiently large error-corrected quantum computer existed, Shor's algorithm would create a path from an exposed secp256k1 public key to its private key. This concerns both ECDSA and BIP 340 Schnorr. Resource estimates depend on assumptions about hardware, error rates, and circuits, so this site does not forecast a practical attack date.

Public-key exposure varies by output type, address reuse, and off-chain information such as extended public keys and descriptors; “never spent means safe” is not a general rule. The dedicated article “Can Quantum Computers Break Bitcoin?” separates the algorithms, exposure models, NIST standards, and Draft migration proposals using primary sources.