Bitcoin mining difficulty has fallen 19.9% from its November 2025 peak of 156 trillion to 126.23 trillion as of the July 25, 2026 retarget — the third-deepest decline since dedicated ASIC hardware replaced GPUs. Network hashrate has contracted roughly 12% from its late-2025 peak above 1 zettahash...
"We intend to harness power toward most productive use." — Fred Thiel, CEO, Marathon Digital Holdings
Bitcoin mining difficulty has fallen 19.9% from its November 2025 peak of 156 trillion to 126.23 trillion as of the July 25, 2026 retarget — the third-deepest decline since dedicated ASIC hardware replaced GPUs. Network hashrate has contracted roughly 12% from its late-2025 peak above 1 zettahash per second to approximately 868 EH/s by late July 2026. According to Hashrate Index data reported by Blockspace, difficulty has now declined year over year for only the second time in Bitcoin's 17-year history. The previous occurrence was mid-2021, when China's mining ban removed roughly half the network's computing power overnight.
The contraction is not a temporary disruption. It reflects a structural reallocation of capital, power capacity, and infrastructure away from SHA-256 hashing and toward artificial intelligence and high-performance computing. Publicly listed miners have committed to over $70 billion in cumulative AI and HPC contracts. They sold more than 32,000 BTC in Q1 2026 alone — a single-quarter record that exceeded combined 2025 sales and surpassed the 20,000 BTC liquidated during the Q2 2022 Terra-Luna collapse.
The compound effect of the April 2024 halving (block reward cut from 6.25 BTC to 3.125 BTC), a Bitcoin price decline from $126,000 in October 2025 to approximately $63,000 in August 2026, and rising energy costs has pushed average production costs above market price for a significant share of the fleet. CoinShares estimates 15–20% of the global mining fleet is now operating at a loss.
Bitcoin mining difficulty peaked at approximately 156 trillion in November 2025, when the network briefly exceeded 1 ZH/s. Since then, the trajectory has been persistently downward:
According to Blockspace, the network has spent more time trimming difficulty than adding it in 2026, indicating machines are leaving faster than new capacity can offset departures. Bitcoin Magazine Pro tracked 287 consecutive days of downward hashrate trend as of late July.
Hashrate Index attributed the 2026 contraction to five factors: weak mining economics, AI/HPC expansion, regional outages (Texas power curtailments, Iran-related disruptions), and policy or energy shocks across mining hubs.
The economic pressure on miners is quantifiable. According to CoinShares' Q1 2026 Bitcoin Mining Report, the average production cost per bitcoin for publicly listed miners rose to approximately $79,995 in Q4 2025. Bitcoin has since traded in the $63,000–$70,000 range, implying many miners lose roughly $10,000–$19,000 per BTC mined.
Hashprice — the standard metric for mining revenue per unit of computational power — tells the same story. Revenue per terahash per day has collapsed from approximately $0.12 in early 2024 to $0.055–$0.065 in mid-2026, according to Hashrate Index. On a per-petahash basis, daily revenue hovers around $29/PH/s, a level comparable to the COVID-era trough of 2020.
The compression stems from three simultaneous forces:
The net effect: CoinShares warns that 15–20% of the global bitcoin mining fleet is unprofitable at current hashprice and power-cost levels.
The structural shift in miner strategy is not speculative — it is documented in signed contracts. Publicly listed bitcoin miners have committed to over $70 billion in cumulative AI and high-performance computing agreements. The largest deals include:
Hut 8: In May 2026, Hut 8 signed a 15-year lease for 352 MW of IT capacity with a base-term contract value of approximately $9.8 billion and potential upside to $25.1 billion if all renewal options are exercised. The company holds 597 MW of contracted AI data center capacity worth approximately $16.8 billion in total. Hut 8's total contracted AI portfolio has reached $26.6 billion.
Core Scientific: Core Scientific is providing CoreWeave with 243 MW of compute as of Q1 2026, with the remaining 347 MW scheduled to be online in early 2027. The total agreement exceeds $10 billion. Core Scientific has committed to selling the majority of its Bitcoin holdings by end of 2026 to fund conversion of its 1.2 GW capacity toward AI data center operations.
IREN (formerly Iris Energy): In May 2026, IREN completed a $3 billion convertible notes deal (upsized from $2 billion due to demand) to fund GPU purchases and data center expansion. The company signed a five-year partnership with Microsoft projected to generate $1.94 billion in annualized revenue at 85% project-level EBITDA margins, and a separate five-year AI cloud contract with NVIDIA valued at $3.4 billion.
Marathon Digital (MARA): MARA announced a partnership with Starwood Capital Group targeting approximately 1 GW of near-term IT capacity with a pathway to 2.5 GW. The company is pursuing a $1.5 billion acquisition of the Long Ridge Energy & Power campus in Ohio. Q1 2026 revenue fell 18% year-over-year to $174.6 million. Net loss widened to $1.3 billion.
TeraWulf and Riot Platforms are pursuing similar infrastructure conversions, though with smaller disclosed contract values.
The economics are straightforward. AI data center leases generate stable, long-term revenue at margins that bitcoin mining cannot currently match. Power contracts and data-center sites entering AI leases are locked in for 15–20 years, making the reversion of this capacity back to Bitcoin mining structurally unlikely absent a major repricing of BTC or a collapse in AI compute demand.
Publicly traded miners sold more than 32,000 BTC in Q1 2026. To contextualize:
| Period | BTC Sold by Public Miners | Approximate Value | |--------|---------------------------|-------------------| | Full Year 2025 | < 32,000 BTC | — | | Q2 2022 (Terra-Luna) | ~20,000 BTC | ~$400M | | Q1 2026 | > 32,000 BTC | ~$2.2B |
Individual miner sales in 2026:
The selling is not panic liquidation. It is deliberate capital reallocation. Miners are converting BTC holdings into cash to fund AI infrastructure buildouts, repay debt, and finance GPU purchases. The economic calculus is simple: the marginal dollar invested in AI infrastructure returns more than the marginal dollar left in bitcoin mining at current price levels.
The hashrate decline raises a recurring question about Bitcoin's security model. A lower hashrate reduces the theoretical economic cost of a 51% attack. During the January 2026 trough (663 EH/s), the estimated attack cost fell 30–40% from peak levels.
Mining pool concentration adds a layer of concern. According to multiple analyses, Foundry and AntPool now control over 51% of Bitcoin's hashrate. Geographic concentration compounds this: the United States, China, and Russia together account for approximately 68% of global hashrate.
However, the absolute hashrate level — even at 868 EH/s — remains orders of magnitude higher than the network's historical average. A sustained 51% attack would still require billions of dollars in hardware and energy, and the economic incentive to execute one remains negligible at current market valuations. The risk is theoretical, not imminent. But the trend line warrants monitoring: if hashrate continues declining at its current rate while pool concentration increases, the security margin narrows.
The bitcoin mining industry is undergoing a structural transformation that has no clear historical precedent. Previous difficulty declines — China's 2021 ban, the 2018 bear market — were driven by external shocks or cyclical price movements. The 2026 contraction is different. It is driven by a rational economic calculation: AI compute pays more than SHA-256 hashing at current margins.
The miners are not distressed sellers. They are infrastructure companies repositioning their most valuable asset — cheap, large-scale power — toward a higher-return use case. The $70 billion in signed AI contracts is not speculative; it represents binding revenue commitments from counterparties including Microsoft, Google, NVIDIA, and CoreWeave.
For Bitcoin, the implications are manageable in the near term but warrant structural attention. The network's difficulty adjustment mechanism is functioning as designed, reducing computational requirements to match available hashrate. But the longer-term question is whether the security budget — funded by a diminishing block subsidy and volatile transaction fees — can sustain sufficient hashrate to maintain the network's security guarantees as infrastructure permanently exits the mining ecosystem.
The data does not yet suggest a crisis. It does suggest a structural repricing of what bitcoin mining infrastructure is worth — and the answer, increasingly, is "more as an AI data center."