Publicly traded Bitcoin miners have signed more than $70 billion in AI and high-performance computing (HPC) contracts since late 2024, converting power capacity originally built for SHA-256 hashing into GPU-dense data centers serving Microsoft, Amazon Web Services, Anthropic, and Google-backed Fl...
"Our bitcoin mining is now essentially in runoff. Signing up another 500 MW or 1 GW of deals over the next few years puts us in a position where we're not a $2.5 billion company, we are potentially a $25 to $30 billion company." — Adam Sullivan, CEO, Core Scientific
Publicly traded Bitcoin miners have signed more than $70 billion in AI and high-performance computing (HPC) contracts since late 2024, converting power capacity originally built for SHA-256 hashing into GPU-dense data centers serving Microsoft, Amazon Web Services, Anthropic, and Google-backed Fluidstack. The shift accelerated after the April 2024 halving compressed hashprice to $29/PH/s/day in Q1 2026 — near or below breakeven for most operators — while AI hosting commands $100–$200+ per kilowatt monthly versus $15–$40 for Bitcoin mining.
The consequences are structural. Core Scientific sold $175 million in Bitcoin holdings in March 2026 to fund AI buildout and described its mining operations as "in runoff." Hut 8 holds $26.6 billion in contracted AI data center revenue across 949 MW of committed IT capacity. Bitcoin's network hashrate fell 27% from its November 2025 peak of 1,160 EH/s to 850 EH/s in February 2026, the steepest decline since China's 2021 mining ban. It has since partially recovered to approximately 1,075 EH/s as of mid-October 2026, but the trajectory raises questions about long-term mining economics and network security budgets.
The April 2024 halving cut Bitcoin's block subsidy from 6.25 BTC to 3.125 BTC. Transaction fees did not compensate. According to CoinShares' Q1 2026 mining report, hashprice — the revenue a miner earns per unit of hashrate per day — fell to approximately $29/PH/s/day in Q1 2026, down from $36–$38 in Q4 2025. For context, the breakeven threshold for operators running current-generation ASICs (sub-15 J/TH) at U.S. grid power rates ($0.06–$0.08/kWh) sits near $30–$35/PH/s/day.
The margin compression triggered the first three consecutive negative difficulty adjustments since July 2022. Approximately 252 EH/s of mining capacity went offline, concentrated among operators running older hardware above 25 J/TH efficiency. Bitcoin's network hashrate dropped from a peak of approximately 1,160 EH/s in October–November 2025 to 850 EH/s by early February 2026 — a 27% decline that CoinDesk characterized as the worst drawdown since China's 2021 mining ban.
The hashrate has since partially recovered. As of mid-October 2026, CoinWarz data shows the network at approximately 1,075–1,101 EH/s, with difficulty at 132.72T. The recovery has been driven by more efficient hardware deployment and improved Bitcoin prices (above $76,000 as of August 2026), but the fleet composition has fundamentally changed: much of the new hashrate comes from operators running dual-use facilities that can redirect power between mining and AI workloads based on relative profitability.
Market assessments compiled by industry trackers put total signed AI and HPC contracts by Bitcoin miners above $70 billion as of early 2026. The deals are not speculative memoranda of understanding. They are multi-year, binding lease agreements with hyperscale cloud providers and GPU hosting firms, backed by power purchase agreements and construction timelines.
The concentration is notable. Three companies — Hut 8, Core Scientific, and TeraWulf — account for the majority of contracted value. The counterparties are not crypto-native firms. They include CoreWeave (backed by Nvidia), Microsoft, Amazon Web Services, Google-backed Fluidstack, and Anthropic. The contracts run 12–15 years, locking in revenue streams that dwarf anything available in proof-of-work mining.
Hut 8 has assembled the largest contract portfolio among former Bitcoin miners. The company holds 949 MW of contracted IT capacity across its AI data center campuses, representing approximately $26.6 billion in expected aggregate base-term contract value, according to company filings.
Key deals:
CEO Asher Genoot has described the strategy as "power-first," stating: "We chase megawatts, not chips." Genoot has characterized AI compute as "critical infrastructure like railroads and steel."
Core Scientific has made the most definitive break from Bitcoin mining among major operators. In March 2026, the company sold approximately $175 million in Bitcoin holdings to fund AI infrastructure expansion. CEO Adam Sullivan described the mining business as "essentially in runoff," with certain operations maintained only to satisfy minimum power commitments at legacy sites being converted to AI colocation.
Key metrics:
TeraWulf has differentiated on power source. The company charges a 20–30% premium above market rates for AI hosting powered by nuclear and zero-carbon energy at its Nautilus facility in Pennsylvania. TeraWulf has signed multi-billion-dollar agreements with Fluidstack to provide hundreds of megawatts of compute in New York. The stock gained approximately 85% year-to-date at its 2026 peak, though it has since corrected roughly 50% from those highs.
MARA announced a strategic joint venture with Starwood Digital Ventures to develop AI and HPC infrastructure. The partnership targets more than 1 GW of near-term IT capacity with a pathway to 2.5 GW. MARA retains an option for up to 50% ownership. Shares rose 17% on the announcement in February 2026, though the company retains a larger mining operation than Core Scientific.
Riot has taken a middle path. For the June 2026 quarter, approximately $60.5 million in revenue came from non-mining activities, split between a data center segment and an engineering business. Riot has been described as having "converted itself into a data-center developer that leases capacity to a single anchor tenant and deliberately runs light on Bitcoin."
Bitdeer reported Q2 2026 revenue of $228.8 million (up from $155.6 million in Q2 2025), though it posted a net loss of $92.3 million as AI infrastructure buildout costs outpaced revenue. AI Cloud annualized recurring revenue grew 77% quarter-over-quarter to $76 million. The company began converting its 180 MW Tydal site in Norway from mining to AI workloads in March 2026.
IREN's AI Cloud services revenue surged 94.2% quarter-over-quarter to $33.6 million in Q3 FY2026, though the company missed both revenue and earnings expectations. IREN announced a 5 GW AI expansion pipeline.
The revenue differential between AI hosting and Bitcoin mining explains the pivot. Industry data compiled by Rax Mining shows:
| Metric | Bitcoin Mining | AI Hosting (Bare Metal) | AI Hosting (Managed GPU) | |--------|---------------|------------------------|-------------------------| | Revenue per kW/month | $15–$40 | $80–$120 | $200–$500+ | | Contract duration | None (spot) | 3–15 years | 1–5 years | | Revenue predictability | Volatile | High | High | | Capex per MW | $1–$3M | $8–$15M | $8–$15M |
The crossover point occurs at approximately $0.04/kWh in power costs: above that threshold, AI hosting generates higher risk-adjusted returns than Bitcoin mining. Most U.S. grid-connected miners operate at $0.06–$0.08/kWh, well above the crossover. Global AI compute demand is projected to grow 3–5x between 2026 and 2030, providing sustained demand for converted capacity.
The conversion is not trivial. Enterprise-grade Tier III AI data center buildout requires 9–18 months and $8–$15 million per MW in capital expenditure. Basic bare-metal colocation can be delivered in 3–6 months for smaller 1–3 MW deployments. The capital intensity explains why miners are raising billions in debt financing (Core Scientific's $1 billion Morgan Stanley facility) and selling Bitcoin reserves.
The market has repriced miners as AI infrastructure companies. All ten of the largest publicly traded mining stocks are in positive territory year-to-date as of mid-2026, according to CoinTelegraph, with gains ranging from 5% to 85%. TeraWulf led at approximately 85%, followed by Hut 8 at roughly 67% and Riot Platforms at approximately 46%.
This performance contrasts sharply with Bitcoin itself, which was trading near $76,000–$83,000 through mid-2026 — below its October 2025 highs. According to CoinTelegraph, mining stocks outperformed Bitcoin in 2026, an unusual divergence that reflects the market treating these companies as data center operators rather than Bitcoin proxies.
The repricing has not been linear. TeraWulf fell approximately 50% from its 2026 peak by October, and Hut 8 shares slipped after a Q2 revenue miss despite growing its AI pipeline. The market is demanding execution, not announcements.
The economic migration raises a question the Bitcoin community has debated since the network's inception: what happens to mining security when the block subsidy diminishes and transaction fees do not fill the gap?
The 27% hashrate decline from November 2025 to February 2026 provides a partial answer. The network continued to function — difficulty adjusted downward, blocks continued to be produced, and no successful attacks occurred. But the episode demonstrated that mining capacity is not permanent. It follows economic incentives, and those incentives now point toward AI.
The Bitcoin network consumed approximately 20 GW of continuous power in 2026, according to estimates based on fleet-weighted ASIC efficiency of 28 J/TH. An unknown but growing fraction of that power sits in dual-use facilities where operators can switch between mining and AI hosting. If AI demand continues to offer 3–10x the revenue per kilowatt, the rational economic decision for these operators is to allocate power to AI during periods of low hashprice and return to mining only when Bitcoin price spikes make it temporarily more profitable.
This creates a new dynamic: Bitcoin's hashrate becomes partially contingent on the AI compute market. A downturn in AI infrastructure spending could return capacity to the Bitcoin network; sustained AI demand could permanently reduce it.
CoinShares' Q2 2026 mining report noted that the mining industry is bifurcating into two groups: pure-play miners operating in low-cost jurisdictions (Kazakhstan, Russia, parts of Africa) running sub-15 J/TH hardware, and dual-use operators in the U.S. and Europe who maintain mining as a secondary revenue stream behind AI hosting.
The Bitcoin mining industry built something the AI sector urgently needs: permitted, grid-connected, power-dense facilities in the 100 MW–1 GW range, often in regions with favorable energy costs. The April 2024 halving exposed the fragility of a business model dependent on an ever-shrinking block subsidy and volatile transaction fees. AI hosting offered a way out — higher revenue, longer contracts, less volatility.
The transition is not complete. Many operators maintain mining as a secondary activity, and a sustained Bitcoin price rally above $100,000 could make mining temporarily competitive again at current difficulty levels. But the direction is clear. The largest public miners are raising debt, selling Bitcoin reserves, and restructuring around 12–15 year AI hosting contracts. The economic incentives have shifted, and the capital is following.
For Bitcoin's network security, the implications are not immediately dire but are worth monitoring. The hashrate has recovered from its February 2026 trough, and the network's difficulty adjustment mechanism ensures continued block production regardless of absolute hashrate levels. The longer-term question — whether transaction fees can sustain a security budget adequate for a multi-trillion-dollar asset without the block subsidy — remains unanswered. The miners are not waiting around to find out.