Bitcoin miners are liquidating reserves, dismantling ASIC racks, and converting data centers into AI infrastructure at a pace that is reshaping both industries. CoinShares projects listed miners will derive 70% of revenue from artificial intelligence by end-2026, up from roughly 30% at the start ...
"Bitcoin is no longer a long-term strategic focus." — Hut 8 Corp, Q4 2025 Earnings Call
Bitcoin miners are liquidating reserves, dismantling ASIC racks, and converting data centers into AI infrastructure at a pace that is reshaping both industries. CoinShares projects listed miners will derive 70% of revenue from artificial intelligence by end-2026, up from roughly 30% at the start of the year. Over $70 billion in cumulative AI and high-performance computing contracts have been signed across the public mining sector since 2025.
The economic logic is straightforward. Bitcoin production costs sit near $90,000 per coin. Spot price hovers around $82,000. Margins are negative. AI infrastructure commands $8-15 million per megawatt in build-out cost but delivers structurally higher, contracted returns. TeraWulf's Q1 2026 results confirmed the crossover: HPC lease revenue of $21 million exceeded bitcoin mining income of $13 million for the first time, accounting for 62% of total quarterly revenue.
The hashrate has responded accordingly. Q1 2026 marked the first first-quarter decline since 2020, ending five consecutive years of double-digit growth. Six consecutive difficulty reductions have occurred in 2026. The network is absorbing a meaningful reduction in global compute with zero operational disruption, but the concentration and security implications merit close examination.
The April 2024 halving cut block rewards to 3.125 BTC. Post-halving production costs for listed miners have converged around $90,000 per bitcoin, according to CoinShares' Q1 2026 Mining Report. With bitcoin trading near $82,000 as of May 2026, the average publicly listed miner operates at a loss on a fully loaded cost basis.
The revenue differential between bitcoin mining and AI infrastructure is stark. Bitcoin mining infrastructure costs approximately $700,000 to $1 million per megawatt. AI data center infrastructure costs $8 million to $15 million per megawatt — roughly 10-15x higher — but the returns are contracted, predictable, and denominated in dollars rather than a volatile commodity.
This is not a speculative pivot. It is an arbitrage on existing assets. Bitcoin miners already own the three things AI compute demands: power purchase agreements, cooling infrastructure, and physical real estate in energy-rich jurisdictions. Converting a bitcoin mine to an AI data center is operationally complex but strategically obvious when mining margins are negative and hyperscalers are bidding aggressively for power capacity.
The scale of committed capital is unprecedented for the mining sector. According to CoinShares and company filings, the following contracts represent the largest publicly disclosed AI/HPC agreements:
| Company | Counterparty | Contract Value | Duration | Capacity | |---------|-------------|---------------|----------|----------| | Core Scientific | CoreWeave | $10.2B | 12 years | 590 MW | | TeraWulf | Multiple | $12.8B (contracted) | Various | 400+ MW | | IREN | Microsoft | $9.7B | Multi-year | 200 MW (76,000 NVIDIA GB300 GPUs) | | Hut 8 | Fluidstack (Google-backed) | $7.0B | 15 years | 245 MW |
These four contracts alone total $39.7 billion. Across the broader public mining sector, cumulative announced AI/HPC contracts exceed $70 billion.
The financing structure is aggressive. IREN carries approximately $3.7 billion in convertible notes. Core Scientific sold 2,385 bitcoin for $208.3 million in Q1 2026 to fund construction. Bitdeer fully liquidated its bitcoin reserves to zero in February 2026. The industry is funding AI expansion through a combination of BTC liquidation, convertible debt issuance, and equity dilution.
Q1 2026 earnings across the mining sector reveal a structural shift in revenue composition.
TeraWulf (WULF): Total revenue of $34 million. HPC lease revenue reached $21 million (62% of total), up 117% sequentially from $9.7 million in Q4 2025. Bitcoin mining contributed approximately $13 million. Management stated it "does not plan to put significant additional capital" into mining operations. GAAP net loss of $427.6 million, driven primarily by non-cash items.
Core Scientific (CORZ): Total revenue of $115.2 million, up from $79.5 million in Q1 2025. Colocation revenue surged to $77.5 million from $8.6 million year-over-year. Self-mining revenue fell to $30.1 million from $67.2 million. Net loss of $347.2 million, including $266.5 million in non-cash impairment charges. More than 450 MW of AI capacity anticipated by end of summer 2026, with 590 MW on track by early 2027.
Riot Platforms (RIOT): Revenue of $167.2 million, with $33.2 million from data center operations.
HIVE Digital Technologies: Reported a 219% year-over-year increase in quarterly revenue, driven by growth in its AI/HPC segment. Secured a $30 million agreement to deploy NVIDIA GPUs for enterprise AI cloud clients.
The pattern is consistent: mining revenue is declining while AI/HPC revenue is scaling rapidly. For companies with active AI contracts, CoinShares projects mining will account for less than 20% of total revenue by end-2026, down from approximately 85% in early 2025.
The network hashrate tells the story numerically. The seven-day simple moving average sits near 965 EH/s as of early May 2026, with daily readings ranging between 899 and 977 EH/s. Year-to-date, hashrate is down approximately 4%.
The most severe drawdown occurred in late January 2026, when Winter Storm Fern forced large-scale miner curtailment in Texas. Hashrate fell 30-40% to a seven-month low of 663 EH/s from peaks near 1.13 ZH/s. Some U.S. pools lost up to 60% of capacity. According to CryptoQuant, this was the worst drawdown since the 2021 China mining ban.
Bitcoin's difficulty has adjusted downward six times in 2026. The most recent reduction, a 2.3% drop on May 1 at block height 947,520, brought difficulty to 132.47 trillion. A February 2026 adjustment projected a 16-18% difficulty drop, one of the largest recent downward shifts. The next adjustment, estimated for May 15, projects an increase to 137.04 trillion.
CoinShares forecasts the network hashrate could reach 1.8 ZH/s by end-2026 and 2 ZH/s by March 2027, but that projection is contingent on bitcoin recovering to $100,000 — a price level that would restore positive margins for remaining operators.
The concentration question is nuanced. Publicly listed U.S. miners have accounted for over 40% of the global hash rate. As these operators redirect capacity toward AI, the absolute hashrate declines, but geographic distribution may improve. A reduction in U.S.-listed miner dominance could produce a more geographically dispersed network — a net positive for censorship resistance.
Short-term 51% attack feasibility increases mathematically as hashrate drops, but the practical risk remains limited. The cost of acquiring sufficient hardware to execute such an attack still runs into billions of dollars. The difficulty adjustment mechanism — which has functioned without interruption through six consecutive reductions in 2026 — ensures the network maintains its 10-minute block cadence regardless of hashrate fluctuations.
The structural concern is longer-term. If bitcoin remains below miner breakeven costs, the industry's compute capacity continues migrating to AI workloads that cannot be redirected back to SHA-256 hashing on short notice. GPU-based AI infrastructure is architecturally incompatible with ASIC-based bitcoin mining. Once a data center is converted, the transition is effectively permanent.
A striking market divergence has emerged in 2026. Year-to-date, TeraWulf is up approximately 85%, Hut 8 up 67%, and Riot Platforms up 46%. Bitcoin is down roughly 20% over the same period.
The market is pricing mining companies not on their bitcoin production capacity but on their AI infrastructure contracts. The $70 billion in contracted revenue provides a revenue visibility that bitcoin mining — inherently tied to commodity price volatility and halving cycles — cannot match.
According to Bernstein, IREN Limited could eventually "sunset" its Bitcoin mining operations entirely as it repurposes sites for GPU-based workloads. This is not an outlier view. It reflects a sector-wide reassessment of what these companies are and what they are worth.
MARA Holdings expanded through acquiring a 64% stake in French data center firm Exaion, signaling that the geographic diversification of the pivot extends beyond U.S. borders.
The bitcoin mining industry is undergoing a structural transformation that is unlikely to reverse. The economic incentives are unambiguous: negative mining margins, $70 billion in contracted AI revenue, and hyperscaler demand for power infrastructure that miners uniquely possess. The companies that survive will not be bitcoin miners. They will be AI infrastructure operators that happen to mine bitcoin on the side.
For the bitcoin network, the implications are mixed. The difficulty adjustment mechanism continues to function as designed, absorbing hashrate declines without service disruption. Geographic decentralization may improve as U.S.-listed miner dominance wanes. But the permanent conversion of ASIC-compatible facilities to GPU-based AI infrastructure represents a structural reduction in the network's potential compute ceiling — one that only sustained price appreciation above $90,000 can arrest.
The market has already rendered its verdict. Mining stocks are trading on AI multiples, not bitcoin production metrics. The hashrate will follow the capital, and the capital has moved.