Bitcoin's network difficulty has declined 19.9% from its November 2025 peak of 155.97 trillion to 126.23 trillion at its July 2026 trough — the third-deepest decline since dedicated ASIC hardware replaced GPUs and only the second year-over-year difficulty decline in the network's history. Hashrat...
"You get a lot more money per electron if you're doing it for AI than for bitcoin mining. By 2028, you'll either be a power generator, be owned by one, or be partnered with one." — Fred Thiel, CEO, MARA Holdings
Bitcoin's network difficulty has declined 19.9% from its November 2025 peak of 155.97 trillion to 126.23 trillion at its July 2026 trough — the third-deepest decline since dedicated ASIC hardware replaced GPUs and only the second year-over-year difficulty decline in the network's history. Hashrate fell approximately 12% from above 1 zettahash per second to roughly 868 exahashes per second over the same period, driven by a 47% decline in BTC price from its October 2025 high of approximately $122,000 to $63,100 at the end of July 2026.
The downturn triggered record miner BTC liquidations. Publicly listed miners sold more than 32,000 BTC in Q1 2026 alone — exceeding combined full-year 2025 sales and surpassing the 20,000 BTC sold during the 2022 Terra Luna collapse. According to CoinShares, the listed mining sector fell below cash breakeven in aggregate in Q2 2026, with the average pre-tax cash cost to produce one Bitcoin reaching approximately $75,500 while BTC traded at $58,400 at quarter-end. Monthly average hashprice hit an all-time low of $27.7/PH/s/day in June 2026.
The proceeds are not being reinvested in mining. Over $70 billion in cumulative AI and high-performance computing (HPC) contracts have been announced across the public mining sector. Hut 8, Core Scientific, TeraWulf, and MARA Holdings are converting gigawatt-scale electrical capacity into AI data center infrastructure through multi-year enterprise agreements. The mining-to-AI pivot is not a temporary hedge — it is a structural reallocation of energy infrastructure away from proof-of-work.
Bitcoin's difficulty registered nine downward adjustments in the first half of 2026. The June adjustment was -10.09%, the 11th-largest single drop in Bitcoin's history. Three consecutive negative adjustments occurred in Q1 — the first such streak since July 2022.
The timeline of the decline:
| Period | Difficulty | Change | |--------|-----------|--------| | November 2025 (Peak) | 155.97T | — | | March 2026 | ~144T | -7.8% single adjustment | | June 2026 | ~131T | -10.09% single adjustment | | July 2026 (Trough) | 126.23T | -19.1% cumulative from peak |
Block cycle durations stretched to 15.6 days versus the standard 14-day target, with average block times reaching 11.1 minutes against the 10-minute design target. The network was producing fewer blocks per day than protocol designers intended, a direct consequence of hashrate departure.
Network hashrate fell from above 1,160 EH/s in early October 2025 to approximately 868 EH/s by late July 2026 — a sustained, 287-day downtrend. According to CoinShares' Q2 2026 mining report, this constituted the network's first six-month hashrate decline since the 2021 China mining ban.
The economics that forced this exit are straightforward. The April 2024 halving cut the block reward from 6.25 to 3.125 BTC. Per-block revenue at October 2025's price peak was approximately $750,000; by mid-2026 it had fallen to $197,000 — a 74% decline. Transaction fees, which spiked during the 2023-2024 inscription boom, normalized to low-single-digit percentages of total block revenue.
CoinShares' Q1 2026 mining report estimated that 15-20% of older mining rigs were operating at a loss. By Q2, the situation had worsened. Key metrics from CoinShares:
Hardware models rendered uneconomical included the MicroBT WhatsMiner M30 series, the Bitmain Antminer S19 (95-110 TH/s variants), and the Canaan AvalonMiner 1246/1346 series. The breakeven electricity rate for these older machines was approximately 5 cents/kWh — a price point no longer available in most U.S. commercial markets.
According to MARA CEO Fred Thiel, speaking to Blockchainreporter on July 23, a megawatt of power dedicated to Bitcoin mining at current prices generates roughly $50,000 to $70,000 in annual revenue, while the same megawatt serving AI inference workloads generates $300,000 to $500,000.
The profitability squeeze forced record asset sales. Publicly listed miners sold more than 32,000 BTC in Q1 2026, according to CoinShares — a single-quarter record that exceeded combined full-year 2025 miner sales and the approximately 20,000 BTC sold during the 2022 Terra-Luna liquidation cycle.
Individual dispositions:
| Company | BTC Sold | Avg Price | Proceeds | |---------|----------|-----------|----------| | Riot Platforms | 3,778 BTC | $76,626 | $289.5M | | Core Scientific | ~1,900 BTC | ~$92,100 | ~$175M (Jan alone) | | Cango | 2,000 BTC | ~$71,500 | ~$143M (Mar, for loan retirement) |
Aggregate miner reserves declined from 1.86 million BTC at end-2023 to approximately 1.8 million BTC by mid-2026, according to on-chain data compiled by CoinShares. The sales were not distressed fire sales in most cases — they funded operational pivots toward AI infrastructure, where capital expenditure requirements are materially higher. According to Thiel, constructing a Bitcoin mining site costs approximately $1 million per megawatt all-in, while an AI data center site costs $10 to $15 million per megawatt for infrastructure alone, before compute hardware.
The capital from BTC sales and new financing is being deployed into AI and HPC hosting. Over $70 billion in cumulative AI/HPC contracts have been announced across the public mining sector, according to CoinShares.
Major deals by company:
Hut 8: Total contracted AI portfolio of $26.6 billion. The company signed a second 15-year Texas lease for 352 MW on July 20, 2026, sending shares up 11% in a single session. Hut 8 also signed a $7 billion contract with Google-backed Fluidstack for a 245 MW AI data center over 15 years. The stock has quadrupled over 12 months.
Core Scientific: Contracted AI portfolio exceeding $24 billion. The company operates approximately 1.1 GW of total leased capacity, including a 530 MW, 15-year AMD partnership. Core Scientific paid $41.9 million to terminate its agreement with Block's Proto division, cancelling approximately 15 EH/s of next-generation 3nm chip deliveries — a direct trade of future Bitcoin mining capacity for AI hosting flexibility.
TeraWulf: Q1 2026 AI hosting revenue of $21 million surpassed its $13 million mining revenue for the first time, marking the crossover point where AI became the company's primary revenue source.
HIVE Digital: Announced a $2.55 billion Toronto AI "super factory" with capacity for over 100,000 GPUs.
MARA Holdings: Sold $1.1 billion in Bitcoin and cut 15% of staff as part of the infrastructure pivot, according to a March 2026 8-K filing.
The revenue differential explains the speed of transition. According to Thiel's July 2026 estimate, AI inference workloads generate 4-7x more revenue per megawatt than Bitcoin mining at current BTC prices and network difficulty levels.
Keel Infrastructure, formerly Bitfarms, provides the most definitive case study. The company shut down its remaining U.S. Bitcoin mining operations on June 29, 2026, after closing its Moses Lake, Washington facility in April. The company completed its U.S. redomiciliation on April 1 and adopted the Keel Infrastructure name as part of its formal repositioning as a developer of data centers and energy infrastructure for HPC and AI workloads.
Financial results for Q2 2026: $30 million in revenue, down 50% year-on-year. Net loss of $65 million. The company reported $819 million in liquidity. Keel sold 1,085 BTC for approximately $75 million to fund its AI infrastructure buildout.
Keel maintains legacy Bitcoin mining operations in Canada, but the U.S. exit — its largest market — represents a decisive strategic withdrawal from proof-of-work mining. The company now describes itself primarily as digital infrastructure, not as a Bitcoin miner.
The hashrate trough did not hold. Bitcoin's network hashrate rebounded from a triple bottom near 850 EH/s in late July to approximately 915 EH/s by mid-August and 954 EH/s by late September, according to CoinWarz data. On October 5, CoinWarz recorded a single-day reading of 1.16 ZH/s, though daily readings are volatile and the seven-day average remained below 1 ZH/s.
Difficulty adjustments turned positive in Q3. After holding near the floor for most of July, difficulty climbed 0.99% on August 8, 1.31% on September 6, and 4.16% on September 19 — the largest single upward adjustment since June. On October 3, difficulty held essentially flat at 132.76T, approximately 15% below the November 2025 record.
Hashprice improved in parallel. The metric surged 22% over the 30 days ending in early October, from $32.42 to $39.63/PH/s/day, according to Luxor data. BTC's recovery from $58,400 at Q2 close to approximately $86,000 in early October was the primary driver. The hashprice recovery lifted the sector above its aggregate cash breakeven, though margins remain thin for operators running older hardware.
The recovery is uneven. The miners that returned hashrate in Q3 are those with access to sub-4-cent electricity and next-generation hardware (Bitmain S21 and later). The retired fleet — S19 and M30 series — is not coming back at current prices.
Mining equities diverged sharply from Bitcoin in early 2026. A basket of publicly listed mining stocks rose approximately 56% through Q1 2026, according to CoinShares, while Bitcoin itself declined 17% over the same period.
The explanation: investors are no longer pricing these companies as Bitcoin proxies. They are being valued as energy and AI infrastructure operators. Core Scientific's Q2 2026 colocation revenue — derived predominantly from AI hosting — was $136.7 million, representing 83% of its $164.2 million total quarterly revenue. Hut 8 reported Q2 2026 total revenue of $74.9 million, of which $72.5 million (97%) was classified as "Compute revenue" — a category that increasingly encompasses AI hosting.
The market is pricing the megawatt, not the hash. An energy asset with a 15-year enterprise contract from an investment-grade counterparty trades at a fundamentally different multiple than a commoditized mining operation exposed to hashprice volatility and halving cycles.
Top banks have taken notice. According to CryptoRank, major financial institutions are now extending credit to mining companies specifically on the basis of their AI hosting contracts, not their Bitcoin production capacity.
The structural reallocation of energy infrastructure from mining to AI raises questions about long-term Bitcoin network security. Network hashrate in early October 2026 remains approximately 20% below its October 2025 peak despite BTC price recovery to $86,000.
Geographic concentration has intensified. The U.S., China, and Russia collectively control approximately 68% of global hashrate, according to Cambridge Centre for Alternative Finance estimates. The exit of marginal U.S. miners has not been offset by equivalent capacity elsewhere.
The economic model underlying network security — that block rewards plus transaction fees incentivize sufficient hashrate to make attacks prohibitively expensive — faces increasing strain. Block rewards halve every four years. Transaction fees have failed to scale as a replacement revenue source. The next halving, projected for 2028, will further compress miner margins unless BTC price appreciates proportionally.
The difficulty adjustment mechanism ensures blocks continue to be produced on schedule regardless of hashrate level. The security question is whether the lower equilibrium hashrate provides sufficient economic protection against state-level adversaries with access to sovereign electricity rates.
The Bitcoin mining industry is undergoing a structural, not cyclical, transformation. Previous difficulty declines — during the 2018 bear market and the 2021 China ban — were followed by hashrate recoveries as price rebounds restored profitability. The 2026 decline is different. Multi-year enterprise AI contracts, with terms running 10-15 years, permanently lock electrical capacity away from mining. A megawatt committed to a 15-year AI hosting deal cannot pivot back to Bitcoin mining regardless of future BTC price.
The data from CoinShares and company filings indicates that the public mining sector has crossed a structural threshold. Companies that entered 2025 as Bitcoin miners are exiting 2026 as energy and AI infrastructure operators. Mining revenue is becoming a secondary or tertiary revenue line, not the primary business.
For Bitcoin's network, the implications are long-term. The difficulty adjustment mechanism ensures operational continuity regardless of hashrate level. But the economic question — whether reduced hashrate levels provide adequate security as block rewards continue their programmatic decline — remains open. The miners have found a more profitable use for their electrons. The network's security model must now compete with that reality.