Publicly traded Bitcoin miners cut realized hashrate by 21% in the first half of 2026 while signing more than $70 billion in AI and high-performance computing contracts with hyperscalers including Microsoft, Google, Anthropic, and AMD. The result is an industry that no longer resembles its origin...
"We are becoming an AI infrastructure company that happens to have a Bitcoin mining legacy." — Adam Sullivan, CEO, Core Scientific, Q2 2026 Earnings Call
Publicly traded Bitcoin miners cut realized hashrate by 21% in the first half of 2026 while signing more than $70 billion in AI and high-performance computing contracts with hyperscalers including Microsoft, Google, Anthropic, and AMD. The result is an industry that no longer resembles its origin: CoinShares projects AI/HPC services will account for 70% of listed miners' revenue by year-end, up from approximately 30% in Q4 2025. Core Scientific, the sector's largest operator by contracted capacity, reported $136.7 million in colocation revenue in Q2 2026 against $27.5 million from Bitcoin mining — a 5:1 ratio.
The pivot has driven a rare divergence. Mining equities have outperformed Bitcoin by roughly 70% year-to-date, with TeraWulf up 85%, Hut 8 up 67%, and Riot Platforms up 46%, even as Bitcoin itself trades below $65,000, approximately 20% off its January levels. Meanwhile, Bitcoin network difficulty has fallen 14% from its 2026 peak to 127.48 trillion, marking only the second year-over-year decline in the network's history, according to Hashrate Index. The economic logic is straightforward: hash price sits near $30 per PH/s per day, a five-year low, while AI colocation delivers 80–90% operating margins on long-term contracts.
Realized hashrate among a tracked cohort of public Bitcoin miners fell from 368.3 EH/s in Q4 2025 to 319 EH/s in Q2 2026, a 13.4% decline, according to Minerweekly. Excluding Bitdeer — which continued to expand its mining-focused ASIC business — the cohort dropped 21.2%, from 324.6 EH/s to 255.9 EH/s over six months.
The broader Bitcoin network has also contracted. Average hashrate declined 10.6% during the same period, with CoinWarz reporting 949 EH/s as of August 19, below the psychologically significant 1 ZH/s threshold that the network briefly touched in late 2025. Mining difficulty fell to 127.48 trillion, 14% below the 2026 peak, with Hashrate Index noting this represents the second year-over-year decline in Bitcoin's 15-year mining history.
The next difficulty retarget is estimated for August 22, 2026, with CoinWarz projecting a modest +0.45% adjustment to 128.06 trillion — suggesting the network is finding a new, lower equilibrium.
The April 2024 halving reduced block rewards from 6.25 BTC to 3.125 BTC. Two years later, the full economic impact is visible.
Production costs: CoinShares' Q1 2026 mining report placed the weighted average cash cost to produce one Bitcoin among public miners at approximately $79,995 in Q4 2025. With Bitcoin trading near $64,400 as of August 19, the gap between production cost and spot price is negative for operators running older hardware or paying above-market electricity rates.
Hash price: The primary revenue metric for miners — hash price — hit $29 per PH/s per day in Q1 2026, down from $36–$38 in Q4 2025. This represents a five-year low. At this level, any machine less efficient than a Bitmain S19 XP operating on power priced above $0.06/kWh is unprofitable. CoinShares estimates this applies to 15–20% of the global mining fleet.
Fleet efficiency: The silver lining for remaining miners: fleet-wide efficiency improved from 21 W/TH to 18.3 W/TH as older machines were retired. Next-generation ASICs rated below 15 J/TH remain profitable at typical industrial electricity rates. The breakeven electricity price for competitive operations has compressed to below $0.05/kWh.
Treasury liquidation: Publicly listed miners have cumulatively reduced BTC treasury holdings by over 15,000 coins from their peak. Core Scientific sold approximately 1,900 BTC in January 2026 alone and signaled plans to liquidate nearly all remaining holdings in Q1 to fund data center conversion.
The aggregate contracted value of AI/HPC deals signed by former Bitcoin miners exceeds $70 billion, according to CoinShares. The counterparties are the largest names in technology:
| Company | Counterparty | Contract Value | Capacity | Duration | |---------|-------------|----------------|----------|----------| | Core Scientific | CoreWeave | ~$10.2B | 590 MW | 12 years | | Core Scientific | AMD | $14B+ | 530 MW (initial) | 15 years | | IREN (Iris Energy) | Microsoft | $9.7B | 200 MW (Phase 1) | 5 years | | IREN | Nvidia | $3.4B | Undisclosed | Undisclosed | | TeraWulf | Fluidstack (Google-backed) | ~$9.5B | 200+ MW | 25 years | | TeraWulf | Core42 | ~$3.3B | 60 MW (energized) | Multi-year | | Hut 8 | Fluidstack/Anthropic | $7.0B | 245 MW (Tranche 1) | 15 years |
Core Scientific's total contracted colocation backlog now exceeds $24 billion following the AMD deal announced in July 2026, which covers up to 2.5 GW of AI infrastructure across multiple sites.
Hut 8's River Bend campus deal with Fluidstack and Anthropic (backstopped by Google) includes options for up to 2,295 MW across three tranches, with the first data hall scheduled for Q2 2027.
The most advanced in its AI transition. Q2 2026 revenue composition: $136.7 million colocation, $27.5 million mining. The company was operating 437 MW of billable AI colocation capacity by mid-July 2026, up from 243 MW at end of Q1. Annualized colocation revenue exceeded $350 million GAAP as of Q1. Total pipeline now targets 2.5 GW. Stock up approximately 40% YTD.
HPC lease revenue surged 117% to $21 million in Q1 2026, surpassing Bitcoin mining income for the first time — HPC accounted for 60% of total revenue of $34 million. The company acquired the Muskie Data Campus in eastern Kentucky and secured agreements with Kentucky Power Company for up to 1 GW of contracted electric service. Stock up approximately 85% YTD — the sector's top performer.
Signed the $9.7 billion Microsoft contract and a separate $3.4 billion Nvidia deal. Microsoft accepted IREN's first AI data center in August 2026 under the contract, prompting a 10% share price jump. Management targets $4 billion in annualized AI cloud run-rate revenue by year-end 2026, with 85% already under contract. The company closed a $3.65 billion GPU-backed financing facility to fund expansion.
Signed the $7 billion River Bend lease in mid-2026. The company is selling $3.25 billion in bonds to finance the Google/Anthropic-backed campus. Development pipeline spans 8.5 GW. Shares initially fell 14.1% on the announcement — the market questioned execution risk on a project that does not begin generating revenue until Q2 2027. Stock has since recovered; up approximately 67% YTD.
Later-stage converts. Riot reported $23.2 million in data center revenue against $113.7 million from mining in its most recent quarter — an 83:17 mining-to-HPC ratio, the inverse of Core Scientific's. Both companies are posting positive YTD stock returns (Riot +46%, MARA +30%) but remain primarily mining operations. The market is pricing in future AI optionality rather than current AI revenue.
The performance gap between mining equities and Bitcoin is the widest on record. Year-to-date through mid-August 2026:
| Asset/Stock | YTD Performance | |------------|----------------| | Bitcoin (BTC) | ~-20% | | TeraWulf (WULF) | +85% | | Hut 8 (HUT) | +67% | | Riot Platforms (RIOT) | +46% | | Core Scientific (CORZ) | +40% | | Applied Digital (APLD) | +37% | | MARA Holdings (MARA) | +30% |
All ten of the largest publicly traded mining stocks are in positive territory YTD, while Bitcoin itself remains down approximately 20%. The correlation between mining stock returns and Bitcoin price, which historically exceeded 0.8, has broken. The market is repricing these companies as AI infrastructure operators, not cryptocurrency proxies.
The migration of industrial-scale power capacity from Bitcoin mining to AI hosting has measurable effects on the Bitcoin network:
Difficulty decline: The 14% drop from 2026's peak difficulty reflects genuine capacity exit, not a seasonal fluctuation. Hashrate Index attributes the contraction to weak mining economics, AI/HPC capital reallocation, Texas power curtailments, and Iran-related disruptions.
Decentralization shift: As public miners reduce hashrate, their share of total network computation declines. Private and sovereign miners — particularly in Russia, Kazakhstan, and parts of the Middle East — may represent a growing proportion of remaining capacity. Data on this shift is limited due to the opaque nature of these operations.
Security budget: Bitcoin's long-term security model depends on miners earning sufficient revenue (block rewards plus transaction fees) to justify the energy expenditure that secures the network. With block rewards halving every four years and hash price at five-year lows, the network's security budget is under structural pressure. Transaction fees contributed approximately 2–3% of miner revenue in Q2 2026, insufficient to offset declining block rewards.
Energy reallocation: The 949 EH/s network consumed an estimated 16.09 GW of power as of August 1, 2026, according to the Cambridge Bitcoin Electricity Consumption Index — annualized at 141 TWh. Each Bitcoin now requires approximately 858,000 kWh to mine. As efficient miners exit for AI, the remaining fleet's average efficiency could deteriorate, increasing per-coin energy consumption even as total network energy declines.
The Bitcoin mining industry's transformation into AI infrastructure is no longer a thesis — it is reflected in quarterly revenue, contracted backlog, and stock valuations. The companies that secured power capacity and grid interconnection agreements for Bitcoin mining are now monetizing those same assets at margins that mining never offered. Core Scientific's 5:1 colocation-to-mining revenue ratio, TeraWulf's 60% HPC revenue share, and IREN's $9.7 billion Microsoft contract represent a structural revaluation of what these companies are.
The implications for Bitcoin's network are less certain. A 14% difficulty decline and the second year-over-year hashrate contraction in network history suggest that the security budget question — long theoretical — is becoming empirical. The remaining miners are more efficient and more committed, but they are also fewer in number and operating on thinner margins. Whether transaction fees can scale to replace declining block subsidies remains the network's most consequential open question.
For investors, the mining-to-AI pivot has created a new asset class: companies with existing power infrastructure, grid agreements, and operational expertise that are being repriced as data center operators. The 70% performance gap between mining equities and Bitcoin in 2026 reflects this repricing. Whether these companies can execute on $70 billion in contracted commitments — building gigawatt-scale data centers on time and on budget — will determine if the current valuations are justified or if the market has traded one form of optimism for another.