Public bitcoin miners have signed more than $70 billion in aggregate AI data-center contracts since early 2025, according to CoinShares. The shift is rewriting the sector's economics: colocation and high-performance computing (HPC) revenue now exceeds mining revenue at multiple listed operators, ...
"Bitcoin miners, once valued mainly for their hash rates and cryptocurrency prices, are increasingly converting their power-intensive data centers to provide computing capacity to AI companies." — Dan Loeb, Founder, Third Point LLC
Public bitcoin miners have signed more than $70 billion in aggregate AI data-center contracts since early 2025, according to CoinShares. The shift is rewriting the sector's economics: colocation and high-performance computing (HPC) revenue now exceeds mining revenue at multiple listed operators, and at least one — Core Scientific — plans to wind down self-mining entirely by the end of 2026. Realized hashrate among a cohort of public miners fell 13.4% between Q4 2025 and Q2 2026, with the decline reaching 21.2% when excluding Bitdeer.
The revenue gap between mining and AI hosting explains the migration. Bitcoin mining yields roughly $80–$120 per megawatt-hour (MWh) at current prices, while AI inference and training workloads on NVIDIA H200 or B200 clusters generate $1,500–$3,500 per MWh, according to industry estimates compiled by CoinTelegraph. AI compute produces approximately 20x more revenue per megawatt than bitcoin mining. Activist hedge fund Third Point LLC disclosed a four-stock basket across Hut 8, Riot Platforms, Core Scientific, and Applied Digital in August 2026, with its Hut 8 position alone valued at $151.8 million.
The pivot raises a structural question for the Bitcoin network: if listed U.S. miners, which have accounted for more than 40% of global hashrate, continue redirecting power to AI, how does network security recalibrate? The data so far suggests the hashrate decline is manageable — difficulty adjustments are functioning as designed — but the concentration risk shifts from a few large miners to a broader, more geographically distributed set of smaller operators.
The scale of capital commitment flowing from AI hyperscalers to former bitcoin mining companies has no precedent in the sector's 15-year history. Over the course of 2025 and early 2026, listed bitcoin miners signed contracts worth more than $70 billion in aggregate, according to the CoinShares Q1 2026 Bitcoin Mining Report. These are not memoranda of understanding or letters of intent; they are multi-year leases with annual escalators, backed in some cases by investment-grade counterparties.
The five largest disclosed deals:
| Company | Counterparty | Capacity (MW) | Contract Term | Base Contract Value | |---------|-------------|---------------|---------------|-------------------| | Core Scientific | AMD | 530 | 15 years | $14.0B | | TeraWulf | Anthropic | 401 | 20 years | $19.0B | | Hut 8 | Fluidstack / Anthropic | 245 | 15 years | $7.0B | | Riot Platforms | Anthropic | 191 | 20 years | $9.1B | | TeraWulf | (HPC total contracted) | — | — | $12.8B |
Extension options on several of these contracts could increase the aggregate value substantially. Riot's Anthropic deal, for example, includes two five-year extension options that would raise total contract value to $16.1 billion. Hut 8's Fluidstack lease could extend to $17.7 billion if all options are exercised. Google, a subsidiary of Alphabet Inc., is providing a financial backstop on Hut 8's River Bend lease, covering lease payments and related pass-through obligations.
Core Scientific is the furthest along in exiting bitcoin mining. In Q2 2026, colocation revenue reached $136.7 million — 83% of total revenue of $164.2 million, up 109% year-over-year from $78.6 million. Self-mining revenue fell 66% to $21.5 million. The company wrote off $41.9 million to terminate its Block chip deal and confirmed it will not invest in new mining equipment. Management stated only one or two mining sites will remain operational by year-end 2026. The 15-year AMD deal covers 530 MW across five campuses, with estimated capex of approximately $6 billion for the build-out.
Hut 8 reported Q2 2026 revenue of $74.9 million, up 81% year-over-year but missing the $80.9 million consensus estimate by 7.4%. The company has 949 MW of contracted IT capacity with approximately $26.6 billion of expected aggregate base-term contract value. Its River Bend campus lease with Fluidstack carries an expected cumulative net operating income (NOI) of $6.9 billion over the base term, translating to an average annual NOI contribution of $454 million. The initial data hall is scheduled for completion in Q2 2027. Hut 8 raised $3.25 billion in project-level debt at 6.192% to finance the Fluidstack build-out.
Riot reported Q2 2026 revenue of $174.2 million, beating analyst estimates by 14.6%, and topped Q1 revenue estimates by 27.2%. The Rockdale, Texas campus will supply 191 MW to Anthropic under a 20-year lease extending through June 2048. Capacity delivery is staged: 96 MW by December 2027 and the full 191 MW by June 2028. Bloomberg reported Anthropic as the counterparty after Riot initially disclosed only a "leading frontier AI lab."
TeraWulf executed a 401 MW, 20-year lease with Anthropic at its Justified Data campus in Hawesville, Kentucky, generating approximately $19 billion in contracted revenue. The company has energized 102 MW of revenue-generating capacity, with an additional 438 MW targeted for energization by H1 2027. In Q2 2026, TeraWulf reported a $940 million loss as it invested heavily in the AI pivot.
The revenue arbitrage driving the pivot is quantifiable.
Revenue per MWh:
Infrastructure costs per MW:
The upfront cost of AI infrastructure is 8–15x higher than mining infrastructure, but revenue per megawatt is approximately 20x higher. The result is significantly better margins on a per-unit basis once facilities are operational, though the capital requirements are an order of magnitude larger.
The electricity cost to mine one bitcoin was approximately $46,426 as of March 2026, with full production cost (including hardware and overhead) at roughly $58,032. With BTC trading around $78,600 in late August 2026, mining margins remain positive but thin. AI hosting contracts, by contrast, offer 10–20-year revenue visibility with 2.5% annual escalators built into the pricing.
CoinShares projects that listed miners could derive as much as 70% of revenue from AI by year-end 2026, up from approximately 30% in Q1 2026. The trajectory is clear: these are becoming infrastructure companies that happen to still mine bitcoin, rather than miners that dabble in data centers.
The migration of power from mining to AI hosting has measurable effects on the Bitcoin network.
Hashrate trajectory in 2026:
Bitcoin's network difficulty stood at approximately 125.81 trillion as of late August, with the next adjustment estimated for September 5, 2026, projected at -0.94%. The year recorded its second-largest difficulty drop, and Hashrate Index noted that bitcoin difficulty fell year-over-year for only the second time in the network's history.
Publicly listed U.S. miners have accounted for over 40% of global hashrate. A reduction in their mining activity could, counterintuitively, improve network decentralization by distributing hashrate more broadly among smaller, geographically dispersed operators. The 51% attack feasibility increases marginally in absolute terms with lower hashrate, but the more distributed ownership structure may offset this risk.
CoinShares still forecasts hashrate growth to around 1.8 ZH/s by end of 2026, conditional on bitcoin recovering toward $100,000. At current prices near $78,600, that projection appears optimistic.
Multiple factors beyond the AI pivot have contributed to 2026 hashrate volatility: Winter Storm Fern caused widespread U.S. power outages in early 2026; Texas power curtailments and Iran-related disruptions affected mining hubs; and post-halving economics continue to pressure marginal operators.
The market has rewarded the pivot aggressively. Year-to-date 2026 stock performance among major miners:
| Ticker | Company | YTD Gain (approx.) | |--------|---------|-------------------| | RIOT | Riot Platforms | +83% | | HUT | Hut 8 | +72% | | BITF | Bitfarms | +50% | | CORZ | Core Scientific | +31% |
Hut 8 shares have climbed nearly 600% year-over-year, and TeraWulf approximately 800%, reflecting the scale of the re-rating from "crypto miner" to "AI infrastructure provider."
Dan Loeb's Third Point LLC assembled a four-stock position across the sector in its latest 13F filing. The fund increased its Hut 8 stake by 51% to $151.8 million. It also invested $7 million in Riot Platforms, $1.4 million in Core Scientific, and less than $1 million in Applied Digital. The basket approach suggests a sector-level thesis rather than a single-company bet.
Core Scientific trades at approximately 25x sales, making it one of the cheaper names relative to its contracted revenue pipeline. The market appears to price in meaningful execution risk on the multi-year build-outs, which is rational given that delivery timelines extend to 2028 and beyond.
Execution risk. Delivering hundreds of megawatts of AI-grade, liquid-cooled infrastructure on schedule is materially harder than plugging in ASIC miners. Timelines stretch to 2027–2028 for initial capacity and require billions in project-level financing. Construction delays, permitting issues, or supply chain disruptions could compress margins.
Capital intensity. Core Scientific estimates approximately $6 billion in capex for its AMD build-out alone. Hut 8 raised $3.25 billion in debt at 6.192%. These are infrastructure-scale capital requirements that fundamentally change the risk profile of companies that were, until recently, relatively asset-light (by comparison) mining operations.
Counterparty concentration. Anthropic appears across multiple contracts — Riot, TeraWulf, and Hut 8 (via Fluidstack). The AI lab's ability to meet its lease obligations over 15–20-year terms depends on its continued commercial viability. Google's financial backstop on the Hut 8 deal mitigates this risk in that specific case, but the broader concentration is notable.
Bitcoin network effects. If the hashrate decline accelerates beyond what difficulty adjustments can absorb gracefully, network security perception could suffer, potentially affecting BTC price — which in turn affects the residual mining revenue these companies still collect. This creates a reflexive dynamic.
AI demand sustainability. The current AI infrastructure buildout assumes sustained demand growth for training and inference compute. A slowdown in AI spending, model efficiency breakthroughs that reduce compute requirements, or regulatory constraints on AI development could affect long-term contract values.
The bitcoin mining industry's pivot to AI infrastructure is not a narrative shift — it is a capital reallocation at scale. More than $70 billion in contracted revenue, multi-year build timelines, and the planned cessation of mining operations at the sector's largest colocation provider confirm that the transition is structural, not opportunistic.
The economic logic is straightforward: AI compute generates roughly 20x the revenue per megawatt that bitcoin mining does. The companies that control power infrastructure, grid interconnections, and data-center permits — assets that take years to develop — hold value that extends well beyond any single use case. Whether they mine bitcoin or host AI workloads is a question of capital allocation, not identity.
For the Bitcoin network, the implications are measurable but manageable. Difficulty adjustments continue to function as designed. The hashrate decline, while notable, may ultimately distribute mining activity more broadly rather than diminish it permanently. The more pressing question is whether bitcoin's price trajectory can sustain the marginal miners who fill the gap left by companies that have moved on.
These companies are no longer bitcoin miners in any economically meaningful sense. They are power infrastructure operators that happen to sit on top of the cheapest electricity in North America. AI is willing to pay more for that electricity than bitcoin is. The market has noticed.