Proof of work: what the electricity actually buys
Mining converts energy into a cost of rewriting history; the security claim rests on that cost, not on the arithmetic being difficult.
Proof of work secures a ledger by making block production expensive in a way that cannot be faked or shortcut. A miner repeatedly hashes a candidate block with a changing number inside it until the resulting fingerprint falls below a target value; there is no method for this other than trying, and trying consumes electricity and machine time. The security claim that follows is economic rather than mathematical: rewriting recorded history means redoing all that work and then outpacing everyone still building forward.
The search, without the metaphor
A hash function turns any input into a fixed-length output that behaves like a random number, cannot be reversed, and changes completely when one input bit changes. A miner assembles a candidate block, varies a field called the nonce, and hashes the header again. If the output is below the current target the block is valid and can be broadcast; if not, the attempt is discarded and the next one begins.
Two properties of this search matter more than the puzzle imagery usually applied to it. First, it is memoryless: a machine that has been hashing for an hour without success is no closer to a solution than one that just started, because each attempt is an independent draw. Second, verification is trivially fast — anyone can check a claimed solution with a single hash — so the asymmetry between producing and checking is the entire trick. Mining is therefore not a computation that produces anything useful outside the system; the output is proof that a quantity of energy was spent on this specific block, in this specific position in the chain.
Difficulty is a thermostat
Mining difficulty is the parameter that sets the target, and it exists because hash power is not constant. If more machines join, blocks arrive faster; the protocol responds by raising difficulty so that average block time returns to its intended value. Bitcoin recalculates every 2,016 blocks, roughly every two weeks, aiming at ten-minute blocks. Other proof-of-work chains retarget on different schedules, some every block.
The causal direction is worth stating plainly because it is often reversed in commentary. Difficulty follows hash power; it does not drive it. A rising difficulty reading means more machines were hashing during the previous period, and the projected next difficulty change is an extrapolation from how fast recent blocks have arrived relative to schedule. Difficulty is also why block production is smooth over months and lumpy over hours: a chain targeting ten minutes will still produce blocks two minutes apart and then forty minutes apart, because the underlying process is memoryless.
What the spending actually buys
An attacker who wants to erase a recent transaction must build an alternative chain from before it and make that alternative the one nodes select. Under the heaviest-chain rule this means producing more cumulative work than the honest network over the same interval, which requires a comparable or greater share of total hash power for as long as the attack lasts, plus the forgone block rewards from mining honestly instead. This is what a 51 percent attack is: a temporary purchase of majority hash power in order to force a chain reorganization.
What such an attack can do is narrow, and the limits are the part most summaries omit. A majority miner can reverse recent blocks, which allows spending coins twice against a counterparty who accepted too few confirmations, and can refuse to include particular transactions. A majority miner cannot move coins without the corresponding private key, cannot create units beyond the issuance schedule, and cannot change the rules, because every full node checks blocks against its own copy of the rules and discards invalid ones regardless of the work behind them. The attack targets ordering, which is the only thing consensus was ever deciding.
Hash rate is an estimate, not an observation
No one can count hashes. Hash rate is inferred from difficulty and the observed pace of recent blocks, which makes it statistically noisy over short windows: a quiet day and a fast day can differ substantially with no change in the underlying machine fleet. Reading a single day's figure as a real change in mining capacity is one of the most common errors made with this metric. Over weeks, the estimate is informative; over hours, it mostly measures luck.
Related figures behave better. Blocks produced in a day shows whether the chain is running ahead of or behind schedule, and average block time shows the same thing in a different unit. Both matter for anyone estimating how long a given number of confirmations will take.
The energy question, framed accurately
Energy consumption in proof of work is determined by miner revenue, not by transaction count. Miners spend up to the point where the marginal cost of hashing equals the expected marginal reward, so the aggregate power draw tracks the value of the block subsidy plus fees, adjusted for electricity prices and hardware efficiency. Adding transactions to a block does not add hashing; a block that is empty and a block that is full cost the same to find.
This is why per-transaction energy figures are analytically awkward. They divide a cost driven by rewards by a denominator driven by demand, so the number falls when usage rises without anything about the mining industry having changed. The honest framing is that proof of work buys a cost of rewriting history, denominated in energy, and the policy debate is about whether that expenditure is worth what it secures. Both sides of that debate can be described neutrally; this site does not take a position on it.
The other structural fact is that the subsidy declines by design. Bitcoin's halving cuts the new issuance per block on a fixed emission schedule, so the flow paid to miners in native units falls over time while transaction fees make up a growing share. That transition has its own lesson later in this track.
Next, the same security question answered a different way, with a bond instead of a bill. For live network figures, compare hash rate and difficulty on the bitcoin asset page, and see methodology for exactly how hash rate is estimated.