Bitcoin’s Hash Rate Fell 13% This Year. Difficulty Fell 2.5%. Only One of Them Is Measured.

We set out to write about Bitcoin’s rising hash rate. It has not been rising. Over the year to 4 August 2026 the daily estimate went from 1,087.8 exahashes per second to 943.1, a fall of 13.3%.¹ It peaked at 1,305.7 EH/s on 25 October 2025 and bottomed at 706.1 EH/s on 25 January 2026. Pulled 4 August 2026.

Before drawing any conclusion from that, look at the second number in the same dataset. Over twenty-six difficulty adjustments across the same period, difficulty fell 2.5%. Two and a half percent, against thirteen. Both numbers describe the same network over the same year, and they disagree by a factor of five.

One of them is measured and one of them is guessed

Nobody counts hashes. There is no meter on the network. Hash rate is inferred backwards from how quickly blocks arrived relative to the difficulty they had to beat, and block arrival is a random process — a Poisson one, with a long tail. A quiet week produces a low estimate and a lucky week produces a high one, and neither requires a single machine to have moved. That is why the same year contains a reading of 706 EH/s and a reading of 1,306 EH/s: an 85% range that no plausible physical account of the mining industry can explain.

Difficulty is different. It is not an estimate of anything; it is a number the network commits to in consensus every 2,016 blocks, derived from the actual time the previous 2,016 took. It is smoothed, it is late, and it is real. Our view is that the industry has this exactly backwards: it reports the noisy inferred figure to four significant figures on a daily basis and treats the committed one as a technical footnote. If you want to know whether Bitcoin’s security budget grew this year, the honest answer from the honest series is that it fell slightly, by about two and a half percent.

The part that is genuinely remarkable

Bitcoin’s price fell 44.1% over the same twelve months. Mining revenue per coin fell with it, and the block subsidy did not increase to compensate. Difficulty — the closest thing to a measurement of how much hardware stayed switched on — fell 2.5%. Miners absorbed a 44% revenue decline and kept almost all of the machines running. Whatever you think of the industry, that is the datapoint that says something about network security, and it is not the one that gets the headline.

What this does not show

Difficulty is not a perfect proxy for deployed hardware either. It responds with a lag of up to two weeks, it cannot distinguish an efficient machine from an old one, and a fleet that halves its energy use while holding output constant looks identical to one that changed nothing. Hash rate and difficulty both describe computational output, not resilience: they say nothing about where the machines are, who owns them, or how many jurisdictions could switch them off.

The daily series we used is one provider’s estimate, computed one particular way, and other trackers publish different figures for the same days. That is itself part of the point. And a falling hash rate is not automatically bad news — it usually means marginal machines went offline, which is the market working, not the network weakening.

We will re-pull both series on 3 November 2026. The test we would apply to anyone quoting a daily hash rate number is simple: ask what difficulty did over the same window, and see whether the two stories match.