56 EH/s Left the Bitcoin Network. Here Is Who It Pays.

Public Bitcoin miners removed roughly 56 EH/s of capacity from the network in the first half of 2026 — a 15% contraction on their side against a 10% fall network-wide, according to a Miner Weekly analysis published 6 September 2026.

Over the same period, AI and HPC revenue at those same companies rose 52% from Q1 to Q2. At the leading names, AI revenue overtook mining revenue for the first time.

The machines did not fail. They were switched off on purpose, because the megawatts behind them were worth more doing something else.

Why public miners cut deeper than the network

The 15%-versus-10% split is the detail worth holding onto. Listed miners with balance sheets, shareholders and AI contracts to sign contracted faster than the network as a whole. Private and international operators — the ones without an AI story to tell investors — largely kept hashing.

The capital behind the shift explains the urgency. Fourteen comparable companies spent $18.6 billion of capex in a single aligned quarter, and six of the AI infrastructure providers spent nearly 15 times their combined revenue for the period. Building AI capacity costs $8–15 million per megawatt against $0.7–1 million for mining, per BitPlanet Research Lab, 2 September 2026.

What the network did about it, automatically

Bitcoin has one mechanism for exactly this situation, and it needs no committee. When hashrate leaves, blocks slow, and within roughly two weeks difficulty falls to compensate. The same block subsidy is then shared among fewer machines.

This is the counter-intuitive result: hashprice rose about 22% over 30 days while capacity was being switched off. Revenue per unit of work improved precisely because competitors left. Miners who stayed on through the summer are being paid for that decision now.

Who collects the relief — and who does not

Difficulty relief is not distributed evenly. It goes to whoever is still hashing, and it is consumed by whoever pays too much for power. At a hashprice of $39.34 per PH/s per day, the break-even electricity price by efficiency looks like this:

Method: efficiency in J/TH multiplied by 24 gives kilowatt-hours consumed per petahash per day; hashprice divided by that gives the electricity price at which revenue equals energy cost. Energy only — it excludes pool fees, hosting fees, depreciation and downtime.

What would reverse it

Three things, in order of likelihood. A sustained Bitcoin price recovery, which raises hashprice and pulls marginal capacity back on. A wave of efficient hardware landing — the newest flagships run at 8.9 J/TH and ship from November. Or AI demand cooling, which would send those megawatts looking for their previous tenant.

The projected +2.94% retarget on 19 September suggests some capacity is already returning. Blocks have been arriving faster than the ten-minute target, which is the network’s way of saying competition is rebuilding.

Verdict

Verdict: the 2026 contraction is not a crisis of Bitcoin’s security — the difficulty mechanism absorbed 56 EH/s without incident and made the remaining miners better off, which is the design working as intended. It is a repricing of who gets to mine.

The winners are operators with efficient hardware and cheap, fixed electricity who simply stayed on. The losers were legacy fleets on floating rates that could not clear 5.6¢/kWh. If capacity keeps returning at +2.94% a retarget, the relief window narrows — and the only durable advantage left is the one written into your energy contract.

Frequently asked questions