Bitcoin’s network difficulty posted one of its largest downward adjustments of the year. Hashrate is falling sharply as miners power down outdated hardware. Mining is no longer profitable for
Bitcoin’s network just went through one of the sharpest corrections its mining sector has seen this year. Difficulty dropped 10.09% at block 953,568, and hashrate contracted to 892.44 EH/s, a 7.69% slide in a single day.
Source: Blockchain.com Behind those numbers sits a straightforward economic reality: mining profitability fell to a level that no longer covers costs for close to a fifth of active miners. A growing share of them are now redirecting computing capacity toward AI hosting deals instead of continuing to mine bitcoin.
Hashprice, the metric that captures actual miner revenue per unit of computing power, dropped below $30 per petahash per second per day in early 2026 — a five-year low. According to CoinShares’ Q1 2026 Bitcoin Mining Report, that level left roughly 15% to 20% of older mining machines cash-flow negative.
The pressure traced back to a brutal Q4 2025. Bitcoin fell from about $124,500 in early October to around $86,000 by the end of December, even as network hashrate stayed near record highs. JPMorgan reached a similar conclusion from a different angle: the bank placed bitcoin’s all-in production cost at about $78,000, built from electricity, hardware depreciation, and overhead across public miners, and found the coin had traded below that cost for five straight months. Roughly one in five miners, by its count, is underwater.
That persistent gap between spot price and breakeven cost is what’s pushing operators to look for revenue outside pure hashing.
<div style=”overflow-x:auto; margin:20px 0;”>
MetricValueTransactions, last 24h612,329Avg. transactions per hour25,514Avg. transaction value0.1227 BTC ($7,628)Median transaction value0.00063 BTC ($39.4)Avg. transaction fee$0.223 (2.3 sats/vB)Block time10m 59sFee share of block reward0.56%Hashrate892.44 EH/s (-7.69% in 24h)Bitcoin mining difficulty fell 10.09% over the weekend, dropping from 138.96 trillion to 124.93 trillion at block height 953,568, according to Galaxy Research, the research arm of Galaxy Digital. The adjustment ranks as the 11th-largest downward move in the network’s history and the second-largest drop of 2026.
The trigger was a roughly 15% bitcoin price decline in June. Squeezed margins forced some operators to shut off unprofitable hardware, and the difficulty epoch stretched to about 15.6 days against a 14-day target before the network corrected. Galaxy attributes the exodus to the June price slide; whether AI conversions accounted for a meaningful share of the exited hashrate is harder to pin down, since no one publishes that breakdown.
A correction of that size doesn’t happen without a meaningful share of computing power leaving the network. Attaching an exact machine count to it, though, goes beyond what any data provider has actually confirmed.
The difficulty data is the hard signal here. Softer corroboration came from Wu Blockchain, which flagged that the Miner Cycle Stress Composite, tracked by Investemais, fell to a fresh 2026 low and entered its “undervalued” range — territory it previously reached near the bottoms of 2015, 2018, 2020, 2022, and 2024. How much weight the composite deserves is debatable; Investemais doesn’t publish its full methodology, and cycle indicators have a habit of working until they don’t. Still, the last time it behaved this way was the 2015 capitulation, when bitcoin fell from about $300 to $160 in under a week.
A year-long Blockchain.com chart showsa divergence that explains much of what’s unfolding now. Between August and December 2025, network hashrate climbed to a record above 1 zettahash per second — 1,042,233,221 TH/s as of December 10 — even as bitcoin’s price had already dropped to around $90,943 from an October peak near $120,000.
Miners kept adding capacity after the price turned lower because equipment orders and power contracts had been locked in months earlier and couldn’t be reversed on short notice. That lag is exactly what set up the current capitulation: capacity added at higher prices became unsustainable once prices fell further.
Miner ClassStatusOperational Thresholds & TacticsIndustrial / Hyper-EfficientProfitableRun new-generation rigs like the Bitmain S21 XP or S23, operating under 15 J/TH, paired with industrial energy contracts under $0.06/kWh or ultra-cheap sources like flared gas.Mid-Tier Fleet OperatorsAt breakeven / at riskRunning older S19-class hardware, they need electricity below $0.055/kWh just to avoid an outright loss.The AI PivotersHedgedLarge public miners like Core Scientific and Terawulf are pivoting data centers toward AI and HPC hosting, subsidizing their tightening bitcoin mining margins.Standard Home MinersUnprofitableMost pay average residential rates above $0.10/kWh and sit deep in negative territory.The mechanism letting part of the industry dodge capitulation is fairly simple. A miner signs a fixed, dollar-denominated contract with an AI hyperscaler, sometimes running as long as 12 years, and in doing so steps entirely outside hashprice volatility.
The money involved dwarfs anything left in pure mining. Core Scientific is raising $3.3 billion through a junk bond sale to finance its shift toward AI-focused data center operations, building six facilities leased to CoreWeave for 12 years and expected to generate approximately $10 billion in revenue. IREN signed a $3.4 billion, five-year AI cloud contract with NVIDIA for air-cooled Blackwell GPUs, alongside a 5GW strategic partnership that includes NVIDIA’s right to invest up to $2.1 billion in IREN stock. A follow-up agreement with Dell isexpected to lift IREN’s annualized run-rate revenue from $3.7 billion to $4.4 billion once the hardware is commissioned in early 2027.
Part of the network’s hashrate decline, in other words, isn’t idle machines sitting dark. It’s companies physically reallocating floor space and power contracts from bitcoin ASICs to GPU clusters.
The AI pivot looks like an escape from volatility, but it swaps one concentration risk for another.
Core Scientific’s entire $3.3 billion bond raise rests on data centers leased to a single tenant — CoreWeave — under a 12-year agreement covering roughly 590 MW of capacity and up to $8.7 billion in potential revenue. If CoreWeave’s own demand falters, or its financing tightens, Core Scientific’s bondholders and shareholders inherit that problem with no second customer to fall back on. IREN’s structure is similar in shape: its projected $4.4 billion run-rate revenue leans on two counterparties, Microsoft and NVIDIA, with NVIDIA doubling as both customer and prospective shareholder through its right to buy up to 30 million IREN shares.
A miner earning hashprice revenue has thousands of anonymous counterparties by design. A converted AI host often has one or two, bound by contracts whose terms the public rarely sees in full. Whether that trade improves or worsens the risk profile depends entirely on how durable AI compute demand proves over the next decade — a question these 12-year leases assume an answer to, rather than test.
Publicly listed miners have cut their aggregate bitcoin holdings by more than 15,000 BTC from peak levels. Core Scientific sold about 1,900 BTC in January alone and plans to liquidate nearly all remaining holdings; Bitdeer zeroed out its treasury in February; Riot sold 1,818 BTC in December 2025. Treasury sales at that pace reflect an industry no longer able to lean on coin appreciation to offset thin cash flow.
A second revenue line is quietly forming around the grid itself. During periods of peak summer or winter demand, some large operators shut their rigs down entirely and resell their pre-purchased, cheap electricity back to the grid at a steep premium — turning what used to be a pure mining operation into something closer to an energy trading business.
Geography is shifting too. Coinshares’ Q1 2026 report points to Paraguay, Ethiopia, and Oman moving into the top ten mining jurisdictions, supported by projects using underused hydroelectric capacity that brings baseline power costs under $0.03/kWh. Consolidation is accelerating in parallel, with well-capitalized firms absorbing distressed mid-tier operators — often buying facilities specifically for their substation access and energy contracts rather than for the rigs inside them. The hardware gap keeps widening as sub-10 J/TH machines like the Bitmain S23 series and SEALMINER A3 split the network into operators who can survive a prolonged low-hashprice stretch and those running legacy fleets headed toward obsolescence.
CoinShares estimates that listed miners could generate as much as 70% of revenue from AI by year-end, up from around 30% today. The next test comes at the June 28 difficulty retarget: if idled rigs come back online on the price bounce, the capitulation reads as a flush; if they don’t, the capacity left for AI permanently.
The post Bitcoin Miner Capitulation Deepens as AI Contracts Replace Hashrate appeared first on ETHNews.
