Bitcoin's mining industry is undergoing a structural transformation. The weighted average cash cost to produce one bitcoin among publicly listed miners reached $79,995 in Q4 2025, according to CoinShares — while the asset trades near $68,000–$70,000. Miners are losing approximately $19,000 per co...
"We are no longer a Bitcoin company. We are an infrastructure-first owner and developer for HPC/AI data centers across North America." — Ben Gagnon, CEO, Bitfarms (now Keel Infrastructure)
Bitcoin's mining industry is undergoing a structural transformation. The weighted average cash cost to produce one bitcoin among publicly listed miners reached $79,995 in Q4 2025, according to CoinShares — while the asset trades near $68,000–$70,000. Miners are losing approximately $19,000 per coin produced. Network hashrate has fallen 27% from its October 2025 peak of 1,160 EH/s to approximately 850–950 EH/s, and mining difficulty dropped 7.76% on March 21, 2026 — the second-largest negative adjustment of the year.
The response has been uniform across the sector: exit bitcoin mining, enter AI infrastructure. Publicly listed miners have signed over $70 billion in cumulative AI and high-performance computing contracts. CoinShares projects some listed miners will derive up to 70% of total revenue from AI hosting by year-end 2026. More than 15,000 BTC have been sold from public miner treasuries since peak holdings, with Bitdeer reducing its balance to zero and Bitfarms — now rebranding as Keel Infrastructure — declaring it will exit bitcoin mining entirely by 2027. The question is no longer whether miners will pivot, but what happens to Bitcoin's security model when they do.
The April 2024 halving cut Bitcoin's block reward from 6.25 BTC to 3.125 BTC per block. Eighteen months later, the consequences have become acute.
According to CoinShares' Q1 2026 mining report, the weighted average cash cost to mine one bitcoin among publicly listed miners stands at approximately $79,995 — a figure that exceeds Bitcoin's trading price by roughly $11,000–$19,000, depending on the day. The metric accounts for direct electricity, hosting, and operational costs but excludes depreciation and SG&A, meaning all-in production costs are higher still.
Hashprice — the revenue a miner earns per petahash per second per day — collapsed from approximately $55/PH/s/day in Q3 2025 to a post-halving low of $28/PH/s/day on February 23, 2026. As of late March, hashprice has recovered marginally to $33/PH/s/day, a level that leaves roughly 15%–20% of the global mining fleet unprofitable, according to CoinShares estimates.
The difficulty-revenue squeeze has real-world consequences. Network difficulty peaked at 155.97 trillion on October 29, 2025 — the highest in Bitcoin's history. It has since declined through multiple negative adjustments, reaching 133.79 trillion on March 21, 2026, after a 7.76% drop. The adjustment mechanism is functioning as designed: unprofitable miners shut down, hashrate falls, difficulty adjusts downward, and remaining miners become marginally more profitable. But the scale and speed of the current cycle is notable.
Total network hashrate peaked at approximately 1,160 EH/s in October 2025 and has fallen to an estimated 850–950 EH/s range, a decline of roughly 18%–27% from peak. In late January, Winter Storm Fern caused a temporary plunge to 663 EH/s across U.S.-based operations — a 30%–40% single-weekend drop — exposing the geographic concentration of mining capacity.
The pivot is not a hedge. For several major listed miners, it is a full strategic repositioning.
Bitfarms → Keel Infrastructure. On February 6, 2026, CEO Ben Gagnon stated the company is "no longer a Bitcoin company." Shareholders approved re-domiciliation to the United States and a rebrand to Keel Infrastructure, with shares trading under ticker KEEL effective approximately April 1, 2026. The company plans to convert its 18 MW Washington mining facility into a GPU-equipped AI data center by December 2026 and fully exit bitcoin mining by 2027. Gagnon stated during the Q4 earnings call: "In time, we will have no bitcoin." The company holds 1,827 BTC, down from a peak of 3,301 BTC.
Core Scientific (CORZ). CEO Adam Sullivan described bitcoin mining as "essentially in runoff," with certain operations maintained only to satisfy minimum power commitments as legacy sites convert to AI colocation. Core Scientific's 12-year contract with CoreWeave — worth over $10 billion — anchors the pivot, covering approximately 590 MW of infrastructure. Core Scientific sold roughly 1,900 BTC ($175 million) in January 2026 and plans to liquidate substantially all remaining holdings through the year. AI colocation revenue rose 268% year-over-year in Q4 2025.
MARA Holdings (MARA). On February 26, 2026, MARA announced a joint venture with Starwood Capital Group to develop 1 GW of near-term AI-capable data center capacity, with a potential pathway to 2.5 GW. CEO Fred Thiel framed the macro shift: "By 2028, you'll either be a power generator, be owned by one, or be partnered with one." MARA previously liquidated over $400 million in BTC to fund the Starwood partnership.
Riot Platforms (RIOT). Riot signed a 10-year lease with AMD in January 2026 to repurpose 25 MW at its Rockdale, Texas facility for AI inference workloads, with expected contract revenue of $311 million over the initial term ($1 billion with extensions). AMD holds options to expand to 200 MW. Riot owns over 1,100 acres and 1.7 GW of power capacity across Texas. The company sold 1,818 BTC ($162 million) in December 2025.
Bitdeer. Fully liquidated its bitcoin reserves to zero in February 2026, selling 943.1 BTC from reserves plus 184 BTC of monthly production. The company raised $325 million in convertible debt and $43.7 million in equity to fund AI and HPC expansion.
The aggregate value of AI and HPC contracts signed by publicly listed bitcoin miners exceeds $70 billion.
The largest single agreement belongs to IREN (formerly Iris Energy), which signed a $9.7 billion, five-year GPU cloud services contract with Microsoft in November 2025. Under the deal, IREN will deploy NVIDIA GB300 GPUs at its 750 MW campus in Childress, Texas, supporting 200 MW of critical IT load. The contract carries approximately $1.9 billion in annual recurring revenue at an estimated 85% EBITDA margin. IREN purchased approximately $5.8 billion in GPU equipment from Dell Technologies to fulfill the agreement. IREN holds zero BTC.
Core Scientific's CoreWeave contract spans 12 years with two five-year extension options. It features take-or-pay provisions at fixed costs with annual escalators. CoreWeave pays for capex, power, and utilities. Neither party can unilaterally terminate. Average annual revenue run rate: approximately $850 million. Note: Core Scientific shareholders rejected a separate $9 billion all-stock merger with CoreWeave in October 2025, but the underlying hosting contract remains intact.
Riot's AMD deal, at $311 million to $1 billion depending on extensions, is smaller but structurally significant: it represents the company's first non-mining revenue stream and validates the conversion thesis for stranded mining assets.
MARA's Starwood partnership targets 1 GW near-term and 2.5 GW long-term, though specific contract values have not been disclosed at the same granularity.
The "mine and hold" strategy that defined public bitcoin miners from 2020 through mid-2025 is over for most of the sector.
Publicly listed miners have collectively reduced their BTC treasuries by over 15,000 BTC from peak levels. The selling has accelerated in Q1 2026:
| Company | Peak BTC Holdings | Current BTC | Net Reduction | |---------|------------------|-------------|---------------| | Bitdeer | ~2,000 | 0 | ~2,000 | | Bitfarms | 3,301 | 1,827 | 1,474 | | Core Scientific | ~2,000+ | Liquidating | ~2,000+ | | Riot Platforms | ~10,000+ | Reduced | 1,818 (Dec alone) | | TeraWulf | 15 | 15 | 0 | | IREN | 0 | 0 | N/A |
The two channels funding the AI pivot are debt issuance and BTC liquidation. Bitdeer raised $325 million in convertible notes. Core Scientific's CoreWeave contract provides capex funding. MARA structured the Starwood JV to share development costs. In each case, bitcoin is treated as working capital rather than a strategic reserve.
Bitcoin's security model depends on miners allocating computational resources to the network. When miners leave for AI workloads, hashrate declines and the theoretical cost of a 51% attack decreases proportionally.
Several mitigating factors apply. Bitcoin's difficulty adjustment mechanism automatically reduces computational requirements when hashrate falls, maintaining approximately 10-minute block times. The network has never experienced a successful 51% attack. Even at 850 EH/s — down from 1,160 EH/s — the cost of mounting an attack remains prohibitively high.
However, the structural concern is not a single attack but long-term hash rate trends. The United States, China, and Russia collectively control approximately 68% of global hashrate. If U.S.-listed miners — which represent a disproportionate share of identifiable hashrate — continue reallocating capacity to AI, the remaining hash rate concentrates in jurisdictions with less operational transparency.
The dynamic creates a feedback loop. Lower BTC prices reduce mining revenue, accelerating the exit to AI. Fewer miners mean lower difficulty, which temporarily improves margins for remaining operators — but also reduces the network's aggregate security budget. If Bitcoin's price does not recover to levels above production cost, the long-term trajectory points toward a smaller, more geographically concentrated mining ecosystem.
Transaction fees, which represent less than 1% of total block rewards at current levels, are insufficient to offset the revenue gap. Bitcoin generated approximately $115 million in annual fee revenue as of late 2025 — a figure that cannot sustain the $44–60 billion annual mining economy, even at reduced scale.
Not every miner is leaving.
CleanSpark (CLSK) maintains over 13,000 BTC on its balance sheet and treats the asset as productive capital — monetizing output through covered calls and exploring bitcoin-backed credit lines. The company continues to prioritize mining in the near term while gradually building AI exposure.
TeraWulf (WULF) holds just 15 BTC, reflecting minimal emphasis on accumulation, but has maintained a pragmatic posture rather than a hard pivot, preserving balance sheet flexibility.
The distinction between the exiters and the holdouts appears to be operational: companies with newer-generation ASICs, low-cost power purchase agreements, and scale advantages can remain marginally profitable at current prices. Companies with higher cost structures or older hardware face an economic imperative to repurpose infrastructure.
The bitcoin mining industry's transformation into AI infrastructure operators is not a temporary diversification — it is a structural exit for multiple major players. The economics are unambiguous: mining bitcoin costs more than bitcoin is worth, while AI hosting contracts offer multi-year revenue visibility at 85% EBITDA margins.
For Bitcoin the network, the implications are real but not immediate. The difficulty adjustment mechanism provides a self-correcting floor on mining viability, and hashrate remains orders of magnitude beyond any practical attack threshold. The longer-term concern is whether a network whose security budget depends almost entirely on inflationary block rewards can sustain adequate hash rate when the next halving in 2028 cuts rewards to 1.5625 BTC per block.
The foundational question is one of economic sustainability — a theme consistent across the blockchain ecosystem. Bitcoin's annual fee revenue of approximately $115 million cannot fund a $44–60 billion mining economy. The subsidy mechanism (block rewards) is mathematically declining. And the entities that previously absorbed those subsidies are now finding higher-margin uses for the same physical infrastructure.
The miners are not abandoning Bitcoin because they lost faith. They are leaving because the numbers no longer work.