The seven-day average passed 1,000 EH/s for the first time since March 3, and chain data shows difficulty already 5.92% higher since October 16, with another increase due around October 29.
Bitcoin's seven-day average hashrate has climbed back above 1 zettahash per second, its highest level since March 3, according to BlocksBridge Consulting's Miner Weekly newsletter published on October 22. The chain has already registered what that recovery costs the miners who stayed: difficulty rose 5.92% at the October 16 retarget and is on course for another increase at the end of the month.
The level is a marker of how far mining capacity has swung this year. Difficulty had been cut six times by May 1, when MiningPool reported the hashrate slipping back below one zettahash as operators redirected power to computing work. CryptoQuant data cited by Cointelegraph put the network's July low near 726 EH/s.
Difficulty is the part of this that is not an estimate. Network hashrate is inferred from how quickly blocks arrive, so every published figure is one provider's calculation over a chosen window. The retarget is recorded on the chain itself. At block 971,712, timestamped October 16 at 12:37 UTC, difficulty moved from 132,716,002,350,731 to 140,577,680,051,919, an increase of 5.92%. The next adjustment falls at block 973,728. With that epoch 42% complete on October 22, mempool.space's public data implied a further rise of about 4.8% and a retarget around October 29. BlocksBridge put the next increase at 5.16%. Both are projections that move with the pace of blocks, and neither has happened yet.
What makes the recovery awkward is where the capacity came from. Listed miners have spent the year converting power contracts and sites to artificial intelligence and high-performance computing work. In earlier work cited in the same newsletter, BlocksBridge put the group's decline in realized hashrate at 23% while directly reported HPC and AI revenue rose 52% quarter over quarter, and valued the mining hardware taken out of service at roughly $1.5 billion. That last figure assumes a purchase price of $20 per terahash, an assumption the firm states and describes as an order of magnitude rather than a market price.
The hashrate those machines represented did not leave the network for long. "The replacement capacity has to come from elsewhere," Miner Weekly wrote. One identifiable source is Bitdeer. Its September production update, released on October 19, put self-mining capacity at 82.3 EH/s against 79.9 EH/s in August, and co-mining capacity at 28.7 EH/s against 21.6 EH/s. Added together, that is 9.5 EH/s of new capacity in a single month from one operator. A year earlier the company reported 35 EH/s of self-mining capacity and 452 bitcoin mined in the month; in September 2026 it mined 1,355, up from the 1,310 it reported for August.
Secondhand machines are moving as well. Miner Weekly noted Sazmining advertising used Antminer S21 XP units with hosting in Paraguay and Texas. It did not say those units came from public miners, and the hardware's origin is not established.
For operators that kept hashing, the arithmetic is the whole story. Difficulty determines how much hashrate is needed to win a given share of block rewards, so a rise reduces the bitcoin a fixed fleet earns. If BlocksBridge's 5.16% projection holds, difficulty will have risen about 11.4% across the two epochs, and a machine whose own hashrate is unchanged would earn roughly 10% less bitcoin than it did before October 16, before any change in the bitcoin price, power price or pool luck. "That's hardly the breathing room remaining miners might have expected from the AI migration," Miner Weekly wrote.
Contracted capacity has also been written off this year. Core Scientific terminated its equipment agreement with Block and Proto Global in the second quarter and recorded a $41.9 million loss; the agreement covered chips representing about 15 EH/s. The newsletter does not say why the agreement ended. Elsewhere the retreat has been from the service layer rather than the hardware: Binance is closing its mining pool and cloud mining products on November 10, and BitFuFu's purchase of 13,700 miners was still unclosed when it was disclosed this month.
Rising hashrate is not read the same way everywhere. Ki Young Ju, founder and chief executive of CryptoQuant, wrote on X that a higher hashrate signals stronger network security and better fundamentals for bitcoin as a store of value. That is an argument about the network rather than about miner margins, and the two can move in opposite directions: the hashrate that secures the chain is the same hashrate that divides the block rewards.
The next retarget is the test. If the current pace of blocks holds, difficulty adjusts again around October 29 at block 973,728, and the network will be close to a tenth harder to mine than it was three weeks earlier.
