Bitcoin's mining difficulty has fallen 19% from its November 2025 peak — the steepest decline since China's 2021 mining ban eliminated roughly half of global hashrate overnight. This time, no government ordered machines unplugged. Public miners are voluntarily redirecting power, capital, and data...
"The current decline has reached similar severity without any government ban, driven entirely by market forces." — CoinShares, Bitcoin Mining Report Q1 2026
Bitcoin's mining difficulty has fallen 19% from its November 2025 peak — the steepest decline since China's 2021 mining ban eliminated roughly half of global hashrate overnight. This time, no government ordered machines unplugged. Public miners are voluntarily redirecting power, capital, and data center capacity toward artificial intelligence and high-performance computing contracts collectively valued at more than $70 billion.
The sector has bifurcated. Companies with signed AI hosting deals — Core Scientific, TeraWulf, Riot Platforms, Hut 8, CleanSpark, IREN — trade at 12.3× EV/NTM sales. Pure-play miners that remain dependent on block rewards trade at 5.9×. The valuation gap reflects a market judgment: grid connections, not ASICs, are the scarce asset. Meanwhile, Bitcoin's 30-day mean hashrate dropped from 1,108 EH/s in November 2025 to 903 EH/s by August 12, 2026 — a nine-month decline of roughly 19% — and difficulty has recorded only its second sub-zero year-over-year reading in network history.
Public miners sold more than 32,000 BTC in Q1 2026 alone, the largest quarterly treasury liquidation on record, to fund construction of AI-grade data center infrastructure. The economic logic is straightforward: hashprice hovers near $30 per PH/s per day, a five-year low, while AI colocation revenue generates $330 million to $1.75 billion in contracted annual net operating income per facility. The question is no longer whether miners will pivot, but whether Bitcoin's network security can absorb the capacity drain.
The April 2024 halving cut Bitcoin's block reward from 6.25 BTC to 3.125 BTC. Network hashrate continued rising through late 2025, peaking above 1,100 EH/s, compressing hashprice — the dollar revenue per petahash per second per day — to approximately $30 by Q1 2026, according to CoinShares. That figure represents a five-year low.
At $30/PH/day, any ASIC less efficient than a Bitmain Antminer S19 XP running on power priced above $0.06/kWh operates at a loss. CoinShares estimates 15–20% of the global mining fleet is currently unprofitable. The average production cost per bitcoin for public miners reached approximately $37,856 post-halving, up from roughly $16,800 before the reward cut, according to the same report.
Meanwhile, AI infrastructure demand has created a parallel revenue opportunity. Hyperscalers, frontier AI labs, and GPU cloud providers need three things miners already possess: megawatt-scale power capacity, grid interconnection agreements, and physical data center shells. The conversion economics are stark: Core Scientific's colocation gross margin hit 59% in Q2 2026, compared to sub-30% margins on remaining mining operations. CleanSpark's 20-year AI lease at its Sandersville, Georgia campus projects $330 million in annual net operating income on 175 MW of capacity — revenue density that no bitcoin mining operation at current prices can match.
Core Scientific reported Q2 2026 total revenue of $164.2 million, up from $78.6 million in Q2 2025. Colocation revenue reached $136.7 million — a 1,190% year-over-year increase from $10.6 million. Self-mining revenue fell 66% year-over-year to $21.5 million. Colocation accounted for 77% of H1 2026 revenue.
The company has signed 15-year lease agreements with AMD and Neocloud for 529–530 MW, with AMD holding options for up to 2.5 GW. CoreWeave contracts cover approximately 590 MW. As of mid-July, Core Scientific was billing for 437 MW at an average annualized colocation GAAP revenue rate of $635 million. Gross margin on colocation stood at 59%.
Core Scientific sold approximately 1,900 BTC (~$175 million) in January 2026 and planned to liquidate substantially all remaining mining-derived bitcoin holdings during Q1.
Riot reported Q2 2026 total revenue of $174.2 million, up 14% year-over-year. Mining revenue declined to $113.7 million, with direct cost to mine rising to $49,912 per bitcoin. Engineering segment revenue tripled year-over-year to $37.3 million, with gross margin expanding from 7% to over 27%.
Riot secured a $9.1 billion AI data center lease, with two major tenants: AMD and an unnamed frontier AI lab. The company reduced its Bitcoin holdings to 11,380 BTC to fund data center development. Net loss was $237.3 million, driven primarily by $75 million in mark-to-market bitcoin losses and $98 million in depreciation.
TeraWulf reported Q2 2026 total revenue of $44.8 million. HPC lease revenue was $31.9 million (71% of total), up 52% quarter-over-quarter. Digital asset revenue was $12.8 million. The company had 102 MW of HPC capacity online at its Lake Mariner facility by mid-year, with a 401 MW lease signed with Anthropic in Kentucky.
Two miner buildings at Lake Mariner were repurposed or placed out of service to support HPC development during H1 2026.
CleanSpark signed a 20-year triple-net lease at Sandersville, Georgia, with an investment-grade global technology company. The deal is expected to generate approximately $6.6 billion in contracted revenue ($11 billion+ if extension options are exercised) and $330 million in annual net operating income starting in late 2027.
Construction costs are estimated at $1.75 billion to $2.1 billion. CleanSpark plans to use its 13,941 BTC treasury as collateral to avoid shareholder dilution. All long-lead items have been ordered and pre-paid, according to the company.
Hut 8's total contracted portfolio reached $26.6 billion in AI/HPC deals, with expected annual net operating income above $1.75 billion. The stock jumped 10% on news of a $9.8 billion AI data center lease.
IREN raised its AI cloud revenue target after signing $2.8 billion in new multi-year contracts. Its year-end 2026 annualized run-rate revenue target for AI cloud services increased to more than $4 billion, up from $3.7 billion, with roughly 85% of that target already under contract. Customers include Microsoft, Nvidia, Perplexity, and Figure AI. S&P Global projects IREN's HPC revenue at 71% of total revenue by year-end 2026, up from 3% in 2024.
Marathon is pursuing conversion through a Starwood partnership, converting roughly 90% of non-hosted mining capacity to AI and critical IT compute sites. However, according to VanEck and multiple analysts, Marathon represents the laggard in HPC colocation, having pursued a "strategic pause" that allowed peers to capture early GPU allocations and primary hyperscaler interest. Marathon's pending acquisition of the 505 MW Long Ridge Energy plant is expected to close in H2 2026.
Bitcoin's 30-day mean hashrate fell from 1,108 EH/s in November 2025 to approximately 903 EH/s by August 12, 2026 — a 19% decline over nine months. The 7-day simple moving average moved from 932 EH/s to 911 EH/s during early August alone.
Mining difficulty peaked at 127.62 trillion and currently sits at 126.23 trillion — roughly 1.1% below the year-ago level. According to CoinWarz data, this is only the second sub-zero year-over-year difficulty reading in Bitcoin's history. The previous instance: China's 2021 ban, which forced an estimated 50% of hashrate offline in weeks.
Public miners specifically reduced realized hashrate by 13.4% between Q4 2025 and Q2 2026, according to Bitbo data. The next difficulty adjustment, estimated around August 23, is projected to decline by another 3.62%.
Bitcoin's Q1 2026 hashrate posted a 4% decline — the first first-quarter drop since 2020. The drop is entirely attributable to capacity reallocation: public miners redirected power toward AI hosting at facilities that previously ran ASICs.
The market has split public miners into two tiers. According to CoinShares and analyst data:
AI/HPC-focused miners (those with signed colocation deals and energized HPC capacity) trade at an average 12.3× EV/NTM sales. This cohort includes Core Scientific, TeraWulf, IREN, Hut 8, and Cipher Mining.
Pure-play mining operators (those still primarily dependent on block reward economics) trade at 5.9× EV/NTM sales.
Year-to-date stock performance reflects the gap. Riot Platforms is up approximately 94% YTD, Hut 8 up 72%, Cipher Mining up 62%, and Core Scientific up 31%, according to market data through mid-August. Analysts have responded in kind: Bernstein raised Core Scientific's target to $32 from $24 and Riot to $30 from $25. B. Riley upgraded TeraWulf, Core Scientific, and CleanSpark. Marathon Digital, by contrast, saw its target cut by multiple coverage initiations.
The underlying valuation thesis has fundamentally shifted. VanEck published a framework arguing that miners should be valued on energized power capacity and tenant quality, not hashrate or bitcoin held. Grid connection agreements — which can take 3–7 years to secure — have become the scarce, non-replicable asset in the portfolio.
The pivot is capital-intensive. According to VanEck analysis, Bitcoin miners pivoting to AI infrastructure face a roughly $50 billion near-term funding gap and as much as $221 billion in long-term capital requirements.
CleanSpark must raise $1.75–$2.1 billion for Sandersville construction. Riot needs to execute on $9.1 billion in contracted leases. Core Scientific must build out up to 2.5 GW of AMD-optioned capacity. TeraWulf must complete its Anthropic facility in Kentucky.
Failure carries significant consequences. CleanSpark's lease agreement includes provisions for rent abatements or lease termination if construction and power integration milestones are not met within 18 months. These are infrastructure projects with multi-year construction timelines, utility interconnection dependencies, and supply chain constraints on electrical switchgear, transformers, and cooling systems.
Funding sources vary. CleanSpark plans to collateralize its 13,941 BTC treasury. Core Scientific and Riot have sold bitcoin directly. Others have tapped convertible debt and equity markets. The $50 billion gap identified by VanEck suggests that not all announced projects will reach commercial operation. Some will be delayed, downsized, or cancelled as capital markets test the sector's execution credibility.
A 19% difficulty decline does not, by itself, compromise Bitcoin's security. The network's difficulty adjustment algorithm — which recalibrates roughly every 2,016 blocks (approximately two weeks) — is specifically designed to accommodate hashrate fluctuations. When hashrate falls, difficulty drops, mining becomes marginally more profitable for remaining participants, and the system reaches a new equilibrium.
However, the structural nature of the current decline differs from historical episodes. Previous large drops (2021 China ban, 2018 bear market) were caused by external shocks or price collapses. The current decline is driven by rational capital allocation: miners are choosing higher-margin AI hosting over lower-margin bitcoin mining. This creates a persistent drag rather than a temporary shock.
Geographic concentration compounds the concern. According to multiple hashrate tracking sources, the United States, China, and Russia together control approximately 68% of global hashrate. As U.S.-based public miners specifically redirect capacity, the relative concentration in other jurisdictions may increase.
The economic cost of a 51% attack remains prohibitive at current hashrate levels — acquiring or deploying 450+ EH/s of mining capacity is not economically rational given the potential reward. But the network's security margin narrows as hashrate falls and consolidates.
The Bitcoin mining industry is undergoing a structural transformation comparable in network impact to China's 2021 ban, but driven by economics rather than regulation. The April 2024 halving compressed mining margins to levels where AI colocation revenue — with gross margins of 50–60% and multi-decade lease terms — represents a categorically superior use of power and data center capacity.
The data is unambiguous. Every major public miner except Marathon has signed billions in AI hosting contracts. Core Scientific's colocation revenue alone exceeds its peak mining revenue. TeraWulf physically repurposed miner buildings for HPC equipment. CleanSpark committed to $6.6 billion in contracted revenue from a single facility. The industry is not diversifying. It is converting.
The open question is whether Bitcoin's network can absorb this drain without material security degradation. The difficulty adjustment mechanism provides short-term resilience. But the miners leaving are not coming back — they have signed 15-to-20-year leases with AMD, Microsoft, Anthropic, and CoreWeave. The megawatts being converted to AI will not run ASICs again. Whatever hashrate level the network stabilizes at will likely reflect a permanent downward repricing of Bitcoin mining's claim on North American power infrastructure.