Public Bitcoin miners have signed more than $70 billion in aggregate AI and high-performance computing (HPC) contracts since late 2024, triggering the fastest sector-wide business model transformation in crypto history. Mining revenue, which constituted 85% of total revenue for listed operators i...
"We are planning to wind down Bitcoin mining by end of 2026. The economics of AI colocation are simply superior." — Adam Sullivan, CEO, Core Scientific (Q1 2026 Earnings Call)
Public Bitcoin miners have signed more than $70 billion in aggregate AI and high-performance computing (HPC) contracts since late 2024, triggering the fastest sector-wide business model transformation in crypto history. Mining revenue, which constituted 85% of total revenue for listed operators in early 2025, is projected to fall below 20% by year-end 2026 for companies that have secured hyperscaler leases.
The pivot carries measurable consequences for Bitcoin's network. Hashrate recorded its first Q1 decline in six years, dropping from 1.13 ZH/s in January to approximately 901 EH/s as of July 17, 2026. Difficulty adjusted downward by 5% on July 11, 2026, to 127.17T. Hashprice sits at $29/PH/s/day — a level last seen during the 2020 COVID crash — rendering all but the most efficient hardware unprofitable above $0.08/kWh electricity costs.
Despite these fundamentals, mining stocks have massively outperformed both Bitcoin and the S&P 500. Miners with secured HPC contracts trade at 12.3x next-twelve-month sales versus 5.9x for pure-play miners, according to CoinShares Q1 2026 data.
Bitcoin miners collectively signed GPU colocation and cloud service agreements with hyperscalers exceeding $70 billion in aggregate value through H1 2026, according to CoinShares. The contracts share common structural features:
The demand driver is straightforward: AI training and inference require exactly what miners already possess — cheap power purchase agreements (PPAs), grid interconnections, cooling infrastructure, and permitting in power-abundant jurisdictions. The conversion economics favor repurposing: miners report 18–24 month build timelines for AI facilities versus 36–48 months for greenfield data center construction.
| Metric | Value | |--------|-------| | Total CoreWeave Contract | $10.2B over 12 years | | Contracted Capacity | 590 MW across 5 sites | | Energized & Billing (Q1 2026) | 243 MW | | Q1 2026 Colocation Revenue | $350M+ GAAP | | AI Revenue Share | 39% of total | | Total Pipeline | 3.0 GW leasable capacity | | Mining Wind-Down | Planned by end of 2026 |
Core Scientific's Denton, TX campus alone accounts for approximately 260 MW of CoreWeave capacity, with Dalton, GA (~175 MW), Muskogee, OK (~70 MW), Marble, NC (~65 MW), and Austin, TX (~20 MW) comprising the remainder. The company expects more than 450 MW billing by end of summer 2026, with full 590 MW operational by early 2027.
| Metric | Value | |--------|-------| | Anthropic Lease | $19B over 20 years | | Total Contracted HPC Revenue | $12.8B+ | | Justified Data Campus (KY) | 401 MW critical IT load | | Muskie Data Campus | Up to 1 GW (delivery from H2 2028) | | Lake Mariner Campus | ~500 MW near-term, expandable to 750 MW | | AI Revenue Share | 27% of total |
TeraWulf's 20-year lease with Anthropic at its Hawesville, Kentucky campus represents the single largest AI-infrastructure contract signed by any former mining company. The Muskie Data Campus is designed for 1 GW total capacity with phased delivery beginning H2 2028.
| Metric | Value | |--------|-------| | Beacon Point Lease | $9.8B over 15 years | | Beacon Point Capacity | 352 MW (NVIDIA DSX architecture) | | Total Contracted AI Capacity | ~597 MW | | Potential Lease Value with Escalators | $25B+ |
| Metric | Value | |--------|-------| | Microsoft Contract | $9.7B over 5 years | | Childress, TX Campus | 200 MW liquid-cooled NVIDIA GB300 | | AI Revenue Share | 9% (scaling rapidly) | | Target Annualized Revenue | $3.1B by year-end | | Additional Partners | Together AI, Fireworks AI, Fluidstack |
IREN's Microsoft contract covers deployment of NVIDIA GB300 GPUs across 200 MW of liquid-cooled infrastructure. The company targets $3.1 billion in annualized recurring revenue by year-end 2026.
The April 2024 halving reduced block rewards to 3.125 BTC, compressing miner revenue at the protocol level. Combined with the AI-driven hashrate migration, the remaining pure miners face severe margin pressure.
Current conditions (July 2026):
| Metric | Value | |--------|-------| | Network Hashrate | ~901 EH/s | | Hashprice | $29/PH/s/day | | Difficulty | 127.17T (adjusted -5% on July 11) | | Break-even (S21 XP) | ~$0.088/kWh | | Break-even (S23 Hydro) | ~$0.124/kWh | | Efficiency Frontier | Sub-15 J/TH | | Average Production Cost | $19,000+ above spot per BTC (for median miner) |
According to CoinShares, the average all-in production cost per Bitcoin for listed miners exceeded spot price by approximately $19,000 in Q1 2026 when including depreciation and SG&A. Only operators running sub-15 J/TH hardware (S21 series, S23 Hydro) at electricity costs below $0.08/kWh maintain positive margins on mining alone.
Hardware efficiency has improved significantly — from 98 J/TH in 2018 to 9.5 J/TH for hydro-cooled S23-class ASICs — but the improvement cannot offset the combination of halved block rewards and compressed hashprice. The cost of acquiring 1 PH/s of hashrate has halved to approximately $10/TH/s, but revenue per PH/s has declined faster.
The migration of power capacity from mining to AI raises questions about Bitcoin's long-term security budget. Key data points:
The economic barrier to attacking Bitcoin remains prohibitively high in absolute terms. However, the rate of hashrate decline — if sustained — could reduce the cost trajectory. The difficulty adjustment mechanism provides automatic rebalancing, but it operates with a ~2-week lag that creates temporary vulnerability windows during rapid hashrate drops.
Counterargument: Lower hashprice creates opportunity for geographically diverse smaller operators in low-cost jurisdictions (Kazakhstan, Ethiopia, Paraguay) to profitably mine, potentially improving the network's geographic distribution even as total hashrate compresses.
The market's verdict on the AI pivot is unambiguous. Year-to-date performance through mid-July 2026:
| Ticker | YTD Return | AI Strategy | |--------|-----------|-------------| | HUT | +363% | Beacon Point + River Bend AI campuses | | WULF | +269% | Anthropic + multi-campus HPC | | IREN | +121% | Microsoft GB300 deployment | | RIOT | +60% | Mixed strategy, slower pivot | | BTC | -46% | N/A |
Miners with secured HPC contracts now trade at 12.3x NTM sales. Pure-play miners trade at 5.9x. The valuation gap implies the market assigns zero or negative value to Bitcoin mining operations and prices these equities purely on data center fundamentals.
This represents a structural re-rating: investors now evaluate former miners against data center REITs and cloud infrastructure providers (Equinix, Digital Realty, CoreWeave) rather than against Bitcoin itself.
The economic value distribution has shifted dramatically. Under the mining model, value flowed: Electricity Provider → Miner → Bitcoin Network (fees) → Token Holders. Under the AI colocation model:
Revenue per MW comparison:
The 8–12x revenue multiplier per megawatt explains why rational operators are exiting mining. The contracts also provide revenue visibility (12–20 year terms) versus mining's exposure to hashprice volatility, halving cycles, and BTC price fluctuation.
However, AI colocation carries concentration risk: CoreWeave alone anchors >$15 billion of the contract pipeline across multiple mining companies. A CoreWeave credit event or demand slowdown would cascade across the sector.
The Bitcoin mining industry is undergoing an identity crisis resolved by economics. The combination of halved block rewards, compressed hashprice, and hyperscaler demand for power infrastructure has made the pivot from mining to AI colocation not merely attractive but economically necessary for publicly traded operators facing shareholder obligations.
The transformation is largely irreversible. Purpose-built GPU facilities operating under 12–20 year leases will not revert to Bitcoin mining. The network must source its security budget from other participants — smaller private operators, sovereign miners, and jurisdictions where electricity costs make mining profitable at sub-$30 hashprice.
The data suggests Bitcoin's security model is entering a transitional phase: the era of publicly traded companies providing hashrate as a byproduct of seeking Bitcoin exposure is ending. What replaces it — whether geographic diversification of smaller operators or a higher BTC price restoring mining economics — remains an open question.