Bitcoin hashprice climbed to $39.63 per petahash a day on September 6, up 22.24% from about $32.42 a month earlier, according to figures published that day by Bitcoin.com News. On a chart it looks like a turn. Miners had spent the summer watching hashprice, the number that decides whether their machines earn money, grind toward the floor, and now it was rising again. The eighth upward difficulty adjustment of the year landed the same week, a 1.31% bump at block 965664 that pushed difficulty to 127.45 trillion.
The recovery has one problem. The miners have not come back.
Hashrate, the total computing power pointed at the network, sat flat near 934 exahashes a second while hashprice rose. That is roughly 200 EH/s below the peak near 1,157 EH/s the network touched in October 2025. When the price of a block reward improves and no new machines switch on to chase it, the improvement is not coming from the mining business getting healthier. It is coming from somewhere else.
Revenue followed the coin, not the fees
Hashprice is a revenue figure, not a difficulty figure. It measures what a unit of computing power earns in a day, and that earning is almost entirely the block subsidy of 3.125 BTC converted at the current bitcoin price, plus whatever transaction fees riders leave behind. In early September bitcoin was hovering near $79,000, off a drop from about $81,300 the week before. As the price ticked up over the month, the same 3.125 BTC was worth more dollars, and hashprice rose with it.
Fees did almost nothing. Bitcoin.com put fees in blocks over the prior day at 0.43% of the total rewards miners collected. So more than 99% of what secured the chain that day was the subsidy, and the subsidy is fixed in coin terms until the next halving. A miner reading the September hashprice rebound is reading a bitcoin price chart with extra steps. The link between what bitcoin does and what a rig earns has been the through-line of this whole cycle, the same force that has kept mining tied to macro rather than to any fee market. It is the reason a rebound in the coin shows up almost instantly in miner revenue while the deeper metrics lag.
The shift in how bitcoin trades against gold and macro data matters here, because a business whose revenue is a geared bet on one asset's dollar price rises and falls with that price and not much else.
Difficulty is still down on the year
Eight increases sounds like a network expanding. Count the other side. Difficulty has moved through ten downward adjustments in 2026 worth a combined 45.27%, against eight increases worth 33.34%. Net, it has fallen from 146.47 trillion in January to 127.45 trillion, a 13% decline. The all-time high of 155.97 trillion was set back in November 2025 and has not been approached since.
Difficulty tracks how much machinery is actually competing, on a two-week lag, so a falling year-on-year reading is the network telling you that hash power left and did not fully return. That is unusual enough that it has only happened one other time in bitcoin's history. The September uptick is a small correction inside a year that still points down.
| Point in 2026 | Hashprice, $/PH per day |
|---|---|
| Late June low | 27.66 |
| August 1 | 31.70 |
| Around August 7 | 32.42 |
| September 6 | 39.63 |
The forward market did not believe in a recovery
Back on August 1, when CoinDesk reported difficulty had shrunk 14% from the year's high, the forward hashprice market was pricing an average of $31.85 per petahash a day through December. Miners, in other words, were selling their expected revenue for the rest of the year at roughly the level it already sat, betting on no recovery at all. The late-June hashprice low had been $27.66. By September 6 the spot reading was $39.63, well above the curve those same operators had signed off on weeks earlier.
That gap is the interesting part of this rebound. It was not planned for. The people closest to hashprice had hedged against a flat-to-worse second half, and the price bounce blew past their own forecast. A move nobody positioned for is a move that can reverse just as fast, because nothing structural underwrote it.
Where the missing machines went
The power that unplugged during the freeze did not vanish. A chunk of it went to artificial intelligence and high-performance computing, where the same warehouses full of electricity and cooling can earn steadier money than a mining rig chasing a 0.43% fee take. CoinDesk tied the difficulty decline directly to capital, power and operators being diverted toward AI. The economics were stark: one analysis found miners had committed billions to AI buildouts even as those bets ran at a heavy paper loss, because the alternative was worse.
Some of the pain showed up on balance sheets rather than in hashrate. IREN, one of the larger listed operators, wrote down $639 million in mining rigs even while its reported revenue rose, the kind of entry that appears when a company decides a slice of its fleet is not worth running at current prices. Machines that get written off do not come back online the week hashprice ticks up 22%.
So the September reading is real and thin at once. Real, because $39.63 is a better number than $27.66 and every operator still plugged in is earning more per machine than they were in June. Thin, because the improvement rests on bitcoin's dollar price, fees remain a rounding error, and the network's own difficulty says the fleet that left has not returned. The next difficulty adjustment and the September inflation print that will push bitcoin around are worth more attention than the 22% headline, which measures the coin as much as the mining.