Bitcoin mining: a margin business with an automatic thermostat

A miner converts electricity into computation and computation into a chance at a block reward. It does not sell bitcoin; it sells a probability, and its share of the total reward is its share of the network's computing power. That framing matters because it explains why a miner's revenue can fall while it works exactly as hard as before - if everyone else works harder, the same machines earn a smaller share.

The industry's unit of revenue is hashprice: the daily revenue per unit of computing power, quoted per petahash per second per day. It captures everything that matters in one number - the coin price, the block subsidy, transaction fees and the total competing hashrate. When hashprice sits near an operator's electricity cost, that operator is at breakeven, and in 2026 that has been a live condition rather than a hypothetical: hashprice reached a low of 27.66 dollars and sat around 32.31 in mid-June, a level widely viewed as near gross breakeven for higher-cost operations.

The block subsidy halves roughly every four years and this is the structural pressure underneath everything. The 2024 halving cut the reward per block from 6.25 to 3.125 coins, halving the industry's revenue overnight at a constant price. The next one arrives in 2028, and the sector is approaching it with visibly thinner margins than it had going into the last one.

Every 2,016 blocks - roughly a fortnight - the protocol adjusts how hard it is to find a block so that blocks keep arriving about every ten minutes. If computing power leaves the network, the task gets easier and the survivors earn more per unit; if it arrives, the task gets harder and everyone earns less. It is a negative feedback loop with no committee, and it makes mining a self-correcting margin business over a two-week cycle.

2026 has been a demonstration of it running downward. A sharp fall in the coin price compressed margins, inefficient machines were switched off, and difficulty fell in three separate adjustments of more than five per cent - including 11.16 per cent in February - with the network's hashrate declining from roughly 1,000 exahashes per second in late 2025 to about 868 by the end of July. Each cut raised the revenue of the machines that stayed on, which is the loop closing.

For anyone trading the coin, the relevant consequence is supply. Miners are structurally sellers - they earn coin and pay costs in currency - and when margins compress they sell more, including from reserves. Over the first half of 2026 miners sold more than 32,000 coins. That is one of the few genuinely observable, recurring sources of supply in this market, which makes it worth watching for the threshold contracts described in the bitcoin entry.

Almost everything here is on-chain and therefore public in near real time. Hashrate is estimated from how quickly blocks are actually found, difficulty is a protocol value visible at every adjustment, and the block subsidy and fee revenue are directly readable. Third parties publish hashprice from those inputs. For a subject with real industrial economics, the transparency is unusual.

One important caveat: hashrate is an estimate, not a measurement. Nobody counts machines. It is inferred from block times, which are random, so short-window hashrate figures are noisy and can move several per cent without any machine changing state. Seven-day averages are the shortest window worth reading, and single-day swings are usually statistics rather than news.

The listed miners publish monthly production updates - coins mined, coins sold, coins held, machines deployed and energised capacity. That monthly cadence is far faster than ordinary corporate reporting and makes the sector's operating condition unusually legible. It is also where the AI pivot becomes visible before it appears in any financial statement.