Bitcoin's network hashrate fell roughly 10% from its October 2025 peak of 1,045 EH/s, marking the first quarterly decline since 2020. Six negative difficulty adjustments in 2026 have brought mining difficulty down 10.7% from its start-of-year peak. The cause is not regulatory pressure or hardware...
"By 2028, you'll either be a power generator, be owned by one, or be partnered with one. The days of being a miner plugged into the grid are numbered." — Fred Thiel, CEO, MARA Holdings
Bitcoin's network hashrate fell roughly 10% from its October 2025 peak of 1,045 EH/s, marking the first quarterly decline since 2020. Six negative difficulty adjustments in 2026 have brought mining difficulty down 10.7% from its start-of-year peak. The cause is not regulatory pressure or hardware failure. It is economic arbitrage: publicly listed miners are rerouting megawatts from SHA-256 hashing to GPU clusters running AI workloads, where revenue per megawatt runs 10x to 20x higher than bitcoin mining.
CoinShares projects 70% of listed miner revenues will derive from AI and high-performance computing by end-2026, up from approximately 30% in Q1. Over $70 billion in cumulative AI/HPC contracts have been announced across the public mining sector. Core Scientific sold $208 million in BTC during Q1 alone. TeraWulf has stated its intention to exit bitcoin mining entirely by year-end. The miners who once hoarded bitcoin are now liquidating it to buy GPUs.
The structural implications extend beyond corporate strategy. With 15-20% of legacy mining rigs operating at a cash loss and hash price at a five-year low of $28-30/PH/s/day, a growing portion of the network's computational security is being economically subsidized by AI revenue from the same facilities. This creates a dependency loop that did not exist 18 months ago.
Bitcoin's network hashrate peaked above 1,045 EH/s in October 2025 following a +6.31% difficulty adjustment that brought mining difficulty to 155.97 trillion. Since then, the trajectory has reversed.
By early February 2026, hashrate had bottomed near 850 EH/s. A partial recovery followed, but Q1 still closed down approximately 4% — the first quarterly decline in six years. As of May 22, 2026, network hashrate sits at roughly 963 EH/s, according to CoinWarz, well below the prior peak.
Six difficulty reductions have occurred in 2026. The most recent cut of 2.3% on May 1 brought difficulty to 132.47 trillion, 10.7% below the January high. Three consecutive negative adjustments earlier in the year matched a pattern not seen since July 2022, a period associated with miner capitulation following the Terra/Luna collapse.
The difference this time: miners are not shutting down. They are redirecting capacity.
The math is straightforward. According to CoinShares' Q1 2026 Bitcoin Mining Report, hash price — the daily USD revenue earned per petahash of capacity — hit roughly $28/PH/s/day in late February before stabilizing in the $30-35 range. This represents a five-year low. The July 2025 peak was $63/PH/s/day.
The weighted average cash cost to produce one bitcoin among publicly listed miners rose to approximately $79,995 in Q4 2025, according to CoinShares. With bitcoin trading around $74,300 in late May, the margin is negative for many operators. CoinShares estimates 15-20% of miners are operating at a cash loss, with mid-generation Antminer S19 fleets requiring sub-5¢/kWh electricity to break even.
The comparative economics of AI hosting are stark:
| Metric | Bitcoin Mining | AI/HPC Hosting | |--------|---------------|----------------| | Revenue per MW | ~$1 million/year | $10-20 million/year | | Hash price trend | Five-year low | N/A | | Contract duration | None (spot) | 5-12 years | | Revenue predictability | Volatile | Contracted |
A single megawatt powering H100 or H200 GPU clusters generates 10x to 20x the revenue of the same megawatt running ASICs. When miners already own the power infrastructure, cooling systems, and data center shells, the switching cost is largely the GPU hardware itself — an expense they are financing through bitcoin liquidation and bond issuance.
Core Scientific has executed the most aggressive pivot. The company sold 2,385 BTC for $208.3 million in Q1 2026 to fund capital expenditures for its AI transition. In April, it closed a $3.3 billion junk-bond offering to accelerate the shift. Its 590 MW contract with CoreWeave is projected to generate $10.2 billion in revenue over 12 years. Q1 results reflected the transition: AI colocation revenue surged 268% year-over-year to $77.5 million, while crypto mining revenue fell to $30.1 million. AI has already become Core Scientific's largest business line.
Riot Platforms reported $167.2 million in Q1 revenue, including $33.2 million from its data center segment. AMD exercised an option for an additional 25 MW at Riot's Corsicana, Texas campus, bringing total contracted capacity to 50 MW. The initial 10-year lease is expected to generate approximately $311 million, with extension options bringing potential revenue to $1 billion. AMD holds additional expansion options for up to 200 MW total. Two-thirds of Riot's Corsicana complex has been repurposed for AI and HPC workloads.
MARA Holdings signed a deal with Starwood Capital Group in February 2026 to develop AI data centers at select locations originally built for bitcoin mining. CEO Fred Thiel has stated MARA intends to "harness power toward most productive use," while maintaining that "Bitcoin remains a core pillar of MARA's strategy."
TeraWulf has taken the most definitive stance. Management has explicitly stated its intention to exit bitcoin mining operations entirely by end-2026. Its nuclear-powered facilities provide a clean energy profile that ESG-conscious hyperscaler tenants find attractive. The company has announced $12.8 billion in contracted HPC revenue.
IREN (formerly Iris Energy) secured a five-year partnership with Microsoft projected to generate $1.94 billion in annualized revenue for AI cloud services.
HIVE Digital Technologies was among the earliest movers, beginning its HPC pivot in 2022. The company operates GPU-based infrastructure across Canada and Europe.
The aggregate numbers are significant. Over $70 billion in cumulative AI and HPC contracts have been announced across the listed-miner peer group. The largest include:
These are multi-year, contracted revenue streams — a structural departure from bitcoin mining's spot-revenue model, where income fluctuates daily with hash price and network difficulty.
According to industry estimates, AI and HPC could represent a $40 billion annual revenue opportunity for transformed miners by 2026. CoinShares projects some operators will derive up to 70% of total revenue from non-mining workloads by year-end.
The financing mechanisms are notable. Miners are funding the transition through three primary channels: selling bitcoin reserves (Core Scientific's $208M Q1 sale), issuing high-yield debt (Core Scientific's $3.3B bond offering), and signing long-term hosting contracts that serve as collateral for further borrowing. CoinShares has flagged that debt levels across the sector are reaching historically unprecedented levels.
The migration raises structural questions about Bitcoin's security model.
Geographic concentration remains a concern. The top three countries — the U.S., China, and Russia — control approximately 68% of global hashrate, with the U.S. alone accounting for 37-38% and gaining roughly 2 percentage points of market share quarter-over-quarter, according to CoinShares data.
The difficulty adjustment mechanism is functioning as designed: as hashrate decreases, difficulty drops, making mining more profitable for remaining operators and theoretically attracting new capacity. The next adjustment, estimated for May 29, 2026, is expected to increase difficulty modestly from 136.61T to 136.84T.
However, a subtler dynamic is emerging. Many of the largest mining operations now cross-subsidize their SHA-256 hashing with AI revenue. A facility running at 60% AI and 40% mining can afford to mine bitcoin at a loss on the mining-only P&L because the AI contracts cover overhead. This means a portion of the network's hashrate is economically sustained not by bitcoin's block reward economics, but by hyperscaler contracts with Microsoft, AMD, and CoreWeave.
If AI demand contracts or hyperscaler clients exercise early termination clauses, the subsidy disappears, and the hashrate that was being economically propped up could go offline rapidly.
CoinShares projects hashrate could still reach 1.8 ZH/s by end-2026, but only if bitcoin returns above $100,000 to make standalone mining broadly profitable.
On the decentralization front, a positive development emerged on May 7, 2026: seven of the world's largest bitcoin mining pools — AntPool, Block Inc., F2Pool, Foundry, MARA Foundation, SpiderPool, and DMND — joined the Stratum V2 Working Group. These pools represent nearly 75% of global hashrate.
Stratum V2 allows individual miners to construct their own block templates rather than accepting assignments from pool operators. The protocol uses end-to-end authenticated encryption, reduces bandwidth by 60-70%, and introduces job negotiation capabilities that redistribute transaction-selection power from pools to miners.
According to analysis by the working group, Stratum V2 could deliver up to 7.4% higher profitability through lower latency and better fee capture — a meaningful margin improvement in a hash price environment where single-digit percentage differences determine viability.
The adoption of Stratum V2 does not resolve the hashrate-AI dependency question, but it addresses a parallel centralization risk: the concentration of block construction authority among a small number of pool operators.
Hashrate is declining structurally, not cyclically. The first Q1 drop in six years reflects a permanent reallocation of power infrastructure from mining to AI, not a temporary dip from market conditions.
Revenue arbitrage is 10-20x. AI/HPC hosting generates $10-20 million per megawatt annually versus approximately $1 million for bitcoin mining. The economic incentive to pivot is overwhelming.
$70 billion in AI contracts have been signed. Multi-year, contracted revenue streams are replacing volatile, spot-based mining income across the listed-miner sector.
15-20% of legacy miners are cash-flow negative. Hash price at $28-35/PH/s/day, a five-year low, has rendered a substantial portion of the network unprofitable on a standalone basis.
Cross-subsidization creates a new dependency. Mining operations sustained by AI revenue may be vulnerable to shifts in hyperscaler demand or contract renegotiations.
Stratum V2 adoption by 75% of hashrate addresses pool centralization but does not resolve the economic concentration emerging from the AI pivot.
The bitcoin mining industry is undergoing its most significant structural transformation since the ASIC transition of 2013-2014. The difference is directional: that shift moved computational resources into bitcoin mining. This one is moving them out.
The transition is rational at the operator level. AI hosting offers multi-year contracted revenue at 10-20x the per-megawatt economics of SHA-256 hashing. No fiduciary acting in shareholders' interests would ignore that differential.
The network-level effects are less straightforward. Bitcoin's difficulty adjustment mechanism was designed for fluctuations in mining profitability, not for a structural reallocation of the industrial base to a competing use case. The system is adapting — difficulty has fallen, and remaining miners are more profitable as a result — but the long-term equilibrium depends on variables outside Bitcoin's control, including AI compute demand growth, hyperscaler contract stability, and BTC price trajectory.
CoinShares' projection that hashrate could reach 1.8 ZH/s by end-2026 is contingent on bitcoin trading above $100,000. At current prices near $74,300, with cash production costs approaching $80,000, standalone mining economics do not support that trajectory.
The miners are not leaving because they lost faith in bitcoin. They are leaving because another buyer will pay 10x more for the same megawatt. That is not a narrative. It is arithmetic.