Difficulty has fallen 15% this year, and it cuts both ways

Almost every projection anyone builds for a mining purchase assumes difficulty goes up. It is a reasonable default over a decade. It has been wrong all year.

Bitcoin network difficulty sits at 125.81 trillion as of 23 August 2026, after a −1.31% adjustment. It entered the year around 148.25 trillion. That is a decline of roughly 15% year to date, across ten downward adjustments against seven increases. The current reading is 0.7% above the year's low of 124.93 trillion, set on 13 June. The next retarget is estimated at about −1.32% on 6 September, though an estimate two weeks out is exactly that.

What that does to a machine

Difficulty is the denominator. Your share of what the network pays out is your hashrate divided by everyone else's, so when everyone else's falls, yours rises without you touching anything.

A machine switched on today is earning against a network roughly 15% smaller than the one a machine bought in January was earning against. Same hardware, same power bill, meaningfully more bitcoin per terahash per day. That is not a forecast, it is arithmetic on a number that has already happened.

For anyone modelling a purchase, the practical version is this: if your spreadsheet has difficulty growth set to a positive number by default — and most do, because most templates were written in a decade when it only went up — it has been telling you the wrong thing since January.

Now the part that should temper it

Difficulty did not fall because mining got easier. It fell because roughly 150 EH/s of capacity left the network, and capacity leaves when it stops paying. Bitcoin is about 38.8% below its October 2025 high of just over $126,000, and that is what has been squeezing operators out.

So the two things move together, and they move against each other in your model:

• Lower difficulty means more coins per machine.
• The weak price that caused it means each of those coins is worth less.

You do not get to pick one. A projection that sets difficulty growth to zero or negative because of this year while also setting price appreciation high is modelling two conditions that have not coexisted — it is taking the good half of a bad market and pairing it with the good half of a good one. The result is a number that has never been true and is not going to be.

The useful discipline is to run the pessimistic case with the pessimistic case: flat or falling price with difficulty recovering as capacity comes back, and see whether the machine still pays.

What it does not tell you

It does not tell you difficulty stays here. Sidelined hardware has not been destroyed; most of it is switched off waiting for a price that makes it worth switching on. A recovery in price brings that capacity back, and difficulty with it, faster than new machines could. The floor is soft in one direction only.

It also does not tell you anything about your own position, because the variable that decides whether a machine pays is your power price, and that has nothing to do with the network. Two people buying the same machine this week, one at four cents and one at nine, are in different businesses.

Run it yourself

Every figure above is dated and sourced, and none of them is a projection. What you actually need is your own numbers.

The calculator takes difficulty growth as an input rather than assuming one, plots what a fleet does against simply buying the same dollars of bitcoin on day one, and tells you which period the two cross over in, if they do. Set difficulty growth negative if you think this year continues. Then set it positive and see what survives.

If the underlying question is whether to own machines at all rather than the coin, that argument is set out here, including the conditions under which the answer is no.