What Is Bitcoin Mining? How Hash Power Turns Electricity Into Blocks

Bitcoin mining is usually introduced with a line about computers solving hard maths problems. It is a useful shorthand and a misleading one. Miners are not working out an answer to anything. They are guessing numbers, billions of times a second, and hoping one guess produces an output that happens to fall below a threshold the network set in advance. It is closer to a lottery draw than to a calculation.

That distinction matters, because almost everything odd about mining follows from it: the electricity bill, the hardware arms race, the way the reward halves every four years, and the fact that a warehouse in Texas can switch off for an hour and make more money than it would have by mining. Here is how the machine actually works, with the network's numbers as they stood on 27 October 2026.

The puzzle is a lottery, not a maths problem

Every Bitcoin block carries a header: the protocol version, a link to the previous block, a fingerprint of all the transactions inside it, a timestamp, the current difficulty target, and a free-form number called the nonce. That header is 80 bytes long. Miners hash it with SHA-256 twice and compare the result to the target.

A valid block is one whose double-hash, read as a number, is lower than the target. There is no way to work backwards from a target to a nonce that satisfies it. The only method is to change the nonce or the timestamp and hash again, over and over. Most attempts fail. One succeeds, and the miner who finds it gets to broadcast the block and claim the reward.

The odds are easy to state. Expected hashes per block are roughly the difficulty multiplied by 2 to the power of 32. At the current difficulty of about 140.6 trillion, that is on the order of 6 x 10^23 hashes for each block found — and the whole network, running at roughly 1,020 exahashes per second, gets through that in around ten minutes.

What miners are actually hashing

The header pins the block's contents via a root hash of its transactions, so a miner cannot quietly change a payment after finding a valid nonce: any edit changes the root, which changes the header, which invalidates the hash. That construction is the Merkle tree, and it is the reason a single 32-byte field can commit to thousands of transactions.

It also explains why mining is not a service the network outsources to computers. A miner proposes a block. Nodes check it: the signatures, the amounts, the proof of work, every rule in the protocol. If the block is wrong, it is discarded and the reward is lost. Miners order transactions and pay for the privilege of trying; they do not decide what counts as valid.

Why the network wants wasted work

The apparent waste is the security model. Rewriting history means re-mining every block since the point you want to change, faster than the rest of the network mines new ones. Because each block costs real electricity and hardware, that attack has a running cost rather than a one-off cost, and it has to be paid continuously for as long as the fraud needs to stand.

Proof of work also solves a naming problem. There is no register of who is allowed to propose blocks, so an attacker could otherwise create thousands of fake identities and outvote honest participants. Requiring each identity to burn computation makes that strategy expensive instead of free. The comparison with the alternative mechanism is set out in our explainer on proof of work versus proof of stake.

The thermostat: difficulty and the ten-minute target

Bitcoin aims for a block every ten minutes, and it enforces that with a difficulty adjustment every 2,016 blocks — roughly two weeks. If blocks came fast, difficulty rises; if they came slowly, it falls. The protocol clamps each adjustment to a factor of four in either direction, which prevents a single strange fortnight from producing an extreme swing.

The mechanism has been visibly working through 2026. At the time of writing the chain tip was block 973,316, the difficulty was 140.6 trillion, and the average block interval over the last difficulty period was 9 minutes 25 seconds — slightly fast, which is why the next adjustment, due at block 973,728, was tracking toward an increase of about 6.3%, after a 5.9% rise at the previous reset.

This is the part of mining that rewards patience and punishes enthusiasm. A miner who buys hardware when prices are high and hash rate is flat will watch the difficulty catch up within a few adjustments, and with it the share of blocks their machines win.

What a block actually pays

A block pays two things: new coins, called the subsidy, and the fees attached to the transactions inside it. The subsidy is currently 3.125 BTC and halves every 210,000 blocks. That produces about 450 new BTC a day across the network, and it will fall to 225 at block 1,050,000 — 76,684 blocks away, or roughly 17 months at the current pace. The mechanics of that event are in our Bitcoin halving explainer; just under 20.1 million of the 21 million coins have been mined.

Fees are the second half of the story, and right now they are almost invisible. Across the fifteen most recent blocks, fees averaged 0.0129 BTC — about 0.41% of the block reward — with individual blocks ranging from 0.002 BTC to 0.056 BTC as the mempool stayed quiet and the prevailing rate sat near one satoshi per virtual byte.

At a BTC price of about $80,000, that put network-wide miner revenue near $36 million a day, almost entirely from issuance. The long-term design assumes fees eventually replace the subsidy. The short-term reality is that miners are being paid to secure the chain, not to process a queue.

From laptops to warehouses: the hardware race

Bitcoin was mined on ordinary CPUs in 2009, then on graphics cards from 2010 once someone noticed that GPUs were far better at repeating one hash function. Field-programmable gate arrays followed, and in early 2013 the first application-specific chips shipped — silicon designed to do nothing but SHA-256.

Efficiency is the number that matters, measured in joules per terahash. The Antminer S9, the workhorse of 2016, ran at roughly 100 J/TH. Machines sold today sit near 15 to 20 J/TH, a five-to-sixfold improvement in a decade. That progress is why the network can grow from a few exahashes to over a thousand while total electricity use has grown far more slowly.

The physical plant has changed too. Racks of air-cooled boxes have given way to immersion tanks and direct liquid cooling, because heat density per square metre is now the binding constraint in most facilities. A modern site is judged on megawatts available, cooling efficiency and how fast it can ramp down.

Pools, shares and how miners get paid

A single machine finding a block is now vanishingly unlikely; a large modern bitcoin mining rig might expect one block every several years of solo mining. So miners join pools, which distribute the work and split the reward. Each miner submits hashes that meet an easier target than the network's — shares — and the pool counts them.

Payout schemes differ in who carries the variance. Pay-per-share models pay for every valid share whether or not the pool finds a block; full-pay-per-share adds the transaction fees; PPLNS pays only from blocks the pool actually finds, which smooths the pool's luck but pushes some of it back onto the miner. Fees for running the pool are typically a percentage of the reward.

Concentration matters more than the schemes. Over the past week, 1,084 blocks were found by 26 identified pools. Foundry USA found 27.0%, AntPool 20.1%, F2Pool 17.1%, ViaBTC 8.5%, MARA Pool 5.3%, Luxor 5.0%, SpiderPool 4.8%, SECPOOL 4.2%, OCEAN 2.7% and Braiins 1.7%. The remaining 40 blocks were spread across sixteen smaller pools. Two pools finding nearly half the blocks is a structural fact worth watching, even though individual miners inside a pool can switch away.

The economics that decide who survives

The industry prices a single unit of hash power per day, called the hash price. Divide $36 million of daily revenue by roughly 1,020,000 petahashes per second of network capacity and you get about $35 per petahash per day.

Now compare that to the cost of running the machine. A rig at 17 J/TH consumes 408 kilowatt-hours a day per petahash. At five cents a kilowatt-hour that is about $20; at eight cents it is about $33, which leaves almost nothing once you account for hardware, staff and cooling. This is the whole business in two lines: the gap between hash price and power cost, multiplied by however many megawatts you can buy cheaply.

It explains behaviour that otherwise looks strange. Miners sign long contracts for curtailable power and get paid to stop using it when the grid is stressed. They chase flare gas, hydro spill and other energy that is stranded or would otherwise be flared. They sell machines the moment a newer generation makes the old ones marginal. And when the hash price falls, the least efficient hardware goes dark first, which pulls difficulty down and restores the margin for everyone still running.

The energy argument, with numbers

Bitcoin's electricity use cannot be metered directly, because nobody knows which machines are running where. It is modelled from hash rate and an assumed hardware mix. The most-cited academic estimate is the Cambridge Centre for Alternative Finance's Digital Mining Industry Report, which surveyed 49 mining firms representing 48% of global hash rate and put annual consumption at 138 TWh — about 0.5% of global electricity — with network emissions of 39.8 million tonnes of CO2 equivalent.

The same survey found sustainable sources at 52.4% of the mix, comprising 42.6% renewables and 9.8% nuclear, up from 37.6% in 2022 — a shift driven partly by coal retiring and partly by where new capacity has been built.

Sensitivity to the hardware assumption is easy to show. At 1,020 EH/s, an all-15 J/TH fleet implies about 15.3 GW and 134 TWh a year; at 20 J/TH it implies 20.4 GW and 179 TWh. That arithmetic is mine, not Cambridge's, and it is the reason estimates from different sources disagree by tens of terawatt-hours without either side being dishonest. Treat any single figure — including the low ones — as a model output.

What mining does not do

A few persistent misunderstandings are worth clearing up, because they lead people to the wrong conclusions about risk.

Mining does not create coins out of thin air beyond the schedule. The subsidy is fixed by protocol and halves on a timetable that was set in 2009; no miner, pool or developer can vote to raise it.

A majority miner cannot take coins from your wallet. It can reverse its own recent payments, double-spend against a merchant who accepts unconfirmed transactions, and refuse to include transactions. To steal from an address it would need the private key, which mining does not supply.

Miners do not decide whether your transaction is valid. They choose which valid transactions to include and in what order, and they are paid more for including those that pay more. Enforcement sits with the nodes that check every block, which is why running one is the only way to verify your own balance rather than trusting someone else's report — a distinction our block explorer guide makes concrete.

And the work is not useful in any other sense. The hashes do not advance science, train models or fold proteins. Their only product is the difficulty of undoing the ledger.

How to read mining data like an operator

Five numbers tell you most of what is happening. Hash rate shows how much computing power is pointed at the network. Difficulty converts that into the expected time between blocks. The hash price converts both, plus the coin price, into daily revenue per unit of hash power. Fee revenue as a share of the block reward shows how close the network is to the day the subsidy no longer carries it. And pool concentration shows how few entities would need to coordinate for a censorship regime to become practical.

Watch the direction of travel rather than any single reading. A rising hash rate with a flat price compresses margins until older machines exit; a difficulty drop after a price fall is not a malfunction but the network's cost floor doing its job. The adjustments are slow by design, which is why mining rewards patience and punishes the assumption that today's margin is permanent.

The industry that grew out of that lottery is now a power business with a hardware cycle attached. It buys electricity at wholesale, sells a commodity priced in dollars per petahash, and lives with a difficulty that resets every two weeks. Everything else — the warehouses, the cooling, the contracts with grid operators — follows from the arithmetic of a single guess repeated a sextillion times.