MINING 06/11
The miners are leaving 35 EH/s and counting
IREN and Core Scientific will be out of bitcoin mining by New Year, taking at least 35 EH/s with them. What that does to everyone still plugged in.
Two of the names that defined public bitcoin mining for five years will not be mining it in eight weeks.
IREN says it will "substantially complete" its exit from mining by December 31, 2026. Core Scientific told its Q3 call it will be out by year-end too — a plan the Wall Street Journal had reported last December as a 2028 target. Together with Cipher, TeraWulf and Hut 8, the five largest converts have more than $50 billion in AI contracts, per CryptoQuant, and at least 35 EH/s is leaving the network with them (Trending Topics).
IREN, by the numbers
The quarter to September 30, from the filing:
- Mining revenue $28 million, down 58% on the quarter and 88% on the year (from $233 million)
- 402 BTC mined, against 2,039 a year earlier
- Average hashrate 8.8 EH/s, down from 45.3 EH/s
- Bitcoin held at quarter end: none — it sells daily
- AI cloud revenue about $200 million, up 182% on the quarter, with a $9.7 billion Microsoft contract in the mix
- Total revenue $228 million; net loss $263 million; adjusted EBITDA $66 million
Shares fell 17% on Thursday with the rest of the AI complex. On Friday Bernstein rated the stock Outperform with a $55 target, on the grounds that owned, unencumbered power is the scarce asset. Both things are true: the business is worth more as a landlord, and the market just repriced landlords.
Core Scientific, by the numbers
Q3 revenue $232.9 million, of which mining was $17.5 million — 7.5%. Self-mining revenue went from $57.4 million to $12.3 million in a year and runs at a gross loss; colocation runs at a roughly 60% gross margin. Miners online at the end of September were nearly half the end-2025 count. The pipeline: 590 MW contracted to CoreWeave across six sites, 530 MW to AMD across five campuses from 2027, about 1.1 GW combined and roughly $1.8 billion a year once built.
Hashrate fell below 900 EH/s in August, the deepest drop since China's 2021 ban, and has since regained about 105 EH/s — mostly from private, state-backed and non-US operators. Output at the miners who stayed — MARA, CleanSpark, Riot — is up 16%. The converts' bitcoin production is down 47% year on year. Every exahash that leaves lowers the difficulty for every exahash that stays. The reward does not shrink because the room got emptier.
What this is, and what it isn't
It is not a verdict on bitcoin. It is a verdict on a specific business — publicly listed, debt-financed, US-based, paying grid prices for power — competing for a fixed reward at a hashprice the trade press calls historically low. That business lost to the alternative use of the same buildings. A data centre leased to a hyperscaler for fifteen years is a bond; a mining farm is a call option on a volatile asset. Shareholders chose the bond.
The bond just got marked down too. Since early October: Cipher −64%, IREN −56%, CoreWeave −52%, Hut 8 −50%, Applied Digital −44%, Core Scientific −32%. The AI pivot was sold as the exit from bitcoin volatility. This week it delivered more of it.
Who is left
The network is being handed to operators with cheaper power and no quarterly call: private firms, sovereign-backed projects, and hosted fleets outside the US. For an individual with machines, the maths is the same as it always was — your cost per kilowatt-hour against everyone else's — but the "everyone else" just got smaller, and the ones leaving were the ones paying the most.
That is the clearest signal mining has sent in two years. It is not that the block reward stopped being worth chasing. It is that it stopped being worth chasing at Texas grid prices with Nasdaq-grade overheads. Somewhere with 4-cent power, a 50% drop in a competitor's hashrate is not a crisis.