Bitcoin miners have signed more than $150 billion in multi-year AI and high-performance computing contracts across 7.5 gigawatts of capacity, according to Bernstein's deal tracker as of July 2026. The shift has produced the first quarterly hashrate decline in six years, with the network's 30-day ...
"Access to power is the AI industry's real bottleneck — and Bitcoin miners, who have spent years acquiring and operating large-scale energy infrastructure, are uniquely positioned to fill that gap." — Gautam Chhugani, Analyst, Bernstein
Bitcoin miners have signed more than $150 billion in multi-year AI and high-performance computing contracts across 7.5 gigawatts of capacity, according to Bernstein's deal tracker as of July 2026. The shift has produced the first quarterly hashrate decline in six years, with the network's 30-day average falling 6.3% quarter-over-quarter to approximately 940 EH/s in Q1 2026 — roughly 12% below the December 2025 peak of 1,066 EH/s.
The economics are straightforward. AI cloud workloads generate $1,600 to $4,000 per megawatt-hour in revenue, compared with $80 to $151 for Bitcoin mining. Gross margins on AI hosting run 95% to 98%; Bitcoin mining margins range from 13% to 83% depending on price, difficulty, and electricity cost. With BTC production costs near $80,000 and the spot price well below that level for much of H1 2026, the capital reallocation is rational. Mining stocks have outperformed Bitcoin by 70% year-to-date. But the pivot raises a structural question: what happens to Bitcoin's security budget when its largest infrastructure operators are economically incentivized to point their megawatts elsewhere?
Bernstein's Bitcoin mining industry deal tracker logged a new AI-related agreement every week in July 2026. The cumulative total: more than 7.5 gigawatts of committed AI and HPC capacity, equivalent to $150 billion in multi-year contract value. CoinShares projects that AI and HPC could account for 70% of listed miners' revenue by year-end 2026, up from approximately 30% at the start of the year.
The demand driver is physical: hyperscalers need power, and they need it now. Data center lead times from greenfield development run 3-5 years. Bitcoin miners already control permitted, grid-connected sites with utility-scale power access — assets that took years to assemble. The conversion timeline from mining to AI hosting is 12-18 months, roughly a third of the greenfield alternative.
The deal structures are long-dated. Most contracts run 15-20 years with investment-grade counterparties, producing predictable cash flows that mining's stochastic block rewards cannot match. Several include renewal options that extend total contract value well beyond base terms.
Hut 8 (HUT). On July 21, 2026, Hut 8 signed a second 15-year lease worth $9.8 billion for 352 MW of IT capacity at its Beacon Point campus in Nueces County, Texas. The deal doubles the tenant's contracted capacity at the site to 704 MW. Campus-level base-term contract value now stands at $19.6 billion, with renewal options pushing the figure to $50.2 billion. Total contracted AI data center capacity across Hut 8's portfolio has reached 949 MW, backed by 1,330 MW of utility capacity, with aggregate base-term contract value at $26.6 billion.
TeraWulf (WULF). TeraWulf signed more than $12.8 billion in long-term, credit-enhanced customer contracts through end-2025. The headline deal: a 20-year lease with Anthropic for a Kentucky AI data center campus, projected to generate $19 billion in contract revenue. TeraWulf plans to raise approximately $3.5 billion in debt financing for the build. In Q1 2026, HPC leasing revenue overtook Bitcoin mining income for the first time — $21 million in HPC revenue out of $34 million total.
IREN (IREN). IREN secured a five-year, $9.7 billion AI cloud contract with Microsoft in November 2025, covering NVIDIA GB300 GPU deployments across 200 MW of liquid-cooled capacity at its Childress, Texas campus. The company disclosed $2.8 billion in additional cloud services contracts with AI developers including Together AI, Fireworks AI, and Fluidstack. IREN is currently at 9% AI revenue but is scaling toward $3.1 billion in annualized recurring revenue.
Core Scientific (CORZ). Core Scientific is converting 300 MW at its Pecos, Texas campus from Bitcoin mining to AI colocation, with plans to scale to 1.5 GW of gross power at that site. AI colocation revenue already accounts for 39% of total revenue. The company's agreement with CoreWeave covers 243 MW of compute, with 347 MW scheduled to come online in early 2027. Total agreement value exceeds $10 billion. Core Scientific also announced a 1.5 GW expansion at a second campus in Muskogee, Oklahoma.
MARA Holdings (MARA). On July 20, 2026, MARA struck a deal to acquire more than 1,200 acres in Matagorda County, Texas, with access to up to 1 GW by October 2027 and 2 GW by spring 2028. The deal is structured as up to $600 million in milestone-based payments, developed alongside Starwood Digital Ventures. Once fully built, MARA's total power portfolio would reach 4.8 GW — rivaling some regional utilities.
Bitdeer (BTDR). Bitdeer liquidated its entire Bitcoin treasury — 3,231 BTC worth over $205 million — by February 2026 to fund its AI pivot. The company has maintained zero net Bitcoin holdings since, converting all freshly mined coins to fiat. Combined with a $325 million convertible notes offering and $43.5 million in equity placements, Bitdeer assembled approximately $575 million for data center expansion, deploying NVIDIA GB200 GPU systems.
The gap between mining and AI hosting economics explains the speed of the transition:
| Metric | Bitcoin Mining | AI Cloud Hosting | |---|---|---| | Revenue per MWh | $80 – $151 | $1,600 – $4,000 | | Gross margin | 13% – 83% | 95% – 98% | | Capital cost per MW | ~$1.2 million | ~$40 million | | Contract duration | None (spot) | 15 – 20 years | | Revenue predictability | Stochastic | Contracted |
The capital intensity gap is significant: AI infrastructure costs roughly 33x more per megawatt to build than mining facilities, reflecting requirements for power redundancy, liquid cooling, and high-speed networking. But the revenue premium — 10x to 50x per MWh — more than compensates over the contract duration. According to multiple industry analyses, AI hosting profits run approximately 2.5x mining per MW on a normalized basis.
The weighted average cost to produce one BTC among listed miners sat at roughly $80,000 in Q4 2025. With Bitcoin trading well below that level for significant portions of H1 2026, an estimated 15-20% of the global mining fleet was operating at a loss. The economic incentive to reallocate capital toward AI workloads has been overwhelming.
Bitcoin's network hashrate peaked at 1,160 EH/s in October 2025. By February 2026, it had fallen to approximately 850 EH/s — a decline of roughly 27% from peak. The 30-day average settled around 940 EH/s by end of Q1, marking the first quarterly decline since 2020 after five consecutive years of double-digit growth.
Multiple factors contributed beyond the AI pivot. Winter Storm Fern in early 2026 forced Texas-based miners to curtail operations during extreme cold, and the resulting difficulty adjustment in February dropped approximately 16-18% — one of the largest recent downward shifts.
The structural concern is longer-term. If AI generates more stable, higher-margin revenue than Bitcoin mining, rational miners will continue to reallocate capital. Each reduction in hashrate reduces the network's economic security budget, funded by block rewards and transaction fees. While a short-term increase in 51% attack feasibility exists, Bitcoin's global hashrate distribution and difficulty adjustment mechanism provide resilience. The network self-corrects: lower hashrate means lower difficulty, which improves margins for remaining miners, establishing a new equilibrium.
However, the trend line is clear. The miners with the largest power portfolios — the ones whose hashrate contributions matter most — are the same ones signing multi-billion-dollar AI contracts. The security budget question is no longer theoretical.
On July 14, 2026, New York Governor Kathy Hochul signed an executive order directing state regulators to pause incomplete permit applications for data centers capable of consuming 50 MW or more. The one-year moratorium — the first statewide ban of its kind in the U.S. — halts construction and expansion permits while agencies develop new environmental standards, conduct a Generic Environmental Impact Statement, and design a framework for community benefits.
The order directly targets the type of power-rich sites that Bitcoin miners are converting for AI use. While the immediate impact is confined to New York, the precedent matters. If other states follow, the conversion timeline lengthens and the miners' key competitive advantage — speed to market versus greenfield development — erodes.
Texas, which hosts the majority of the deals catalogued above, has not indicated similar restrictions. But grid reliability concerns from ERCOT during extreme weather events suggest the regulatory conversation is not limited to one state.
Mining stocks have materially outperformed Bitcoin in 2026. According to multiple sources, the sector has beaten BTC by approximately 70% year-to-date. Individual standouts:
The market is pricing these companies as infrastructure operators with contracted revenue streams, not as levered Bitcoin proxies. The re-rating reflects the shift from commodity-price-dependent mining margins to predictable, long-duration cash flows backed by investment-grade tenants.
The Bitcoin mining industry is undergoing a structural transformation, not a cyclical adjustment. The $150 billion AI contract pipeline, 7.5 GW of committed capacity, and first hashrate decline in six years represent a fundamental change in how these companies allocate capital and generate revenue. The economic logic is clear: at current Bitcoin prices and mining costs, AI hosting is the higher-return use of every available megawatt.
The transition is not without risk. Capital costs for AI infrastructure are 33x higher per megawatt. Contract execution over 15-20 year terms requires sustained AI demand growth. Regulatory intervention — as New York demonstrated — could slow the pipeline. And Bitcoin's security budget faces a structural challenge as its largest power consumers redirect capacity.
For the Bitcoin network, the question is whether remaining miners and new entrants will fill the gap at lower difficulty levels, or whether the security budget will settle at a structurally lower equilibrium. The difficulty adjustment mechanism provides a floor, but it does not guarantee the same level of economic security that 1,160 EH/s provided. The market will resolve this, but the resolution may take years.