Bitcoin's network hashrate has spent 316 consecutive days below its October 2025 all-time high of 1,151 EH/s, the longest such drought in a decade. As of early September 2026, the smoothed metric sits near 914 EH/s — a 21% decline from peak. The primary driver is not a market crash or regulatory ...
"We are effectively becoming a data-center company that happens to mine Bitcoin." — Adam Sullivan, CEO, Core Scientific (Q2 2026 earnings call)
Bitcoin's network hashrate has spent 316 consecutive days below its October 2025 all-time high of 1,151 EH/s, the longest such drought in a decade. As of early September 2026, the smoothed metric sits near 914 EH/s — a 21% decline from peak. The primary driver is not a market crash or regulatory crackdown. It is a calculated capital reallocation: publicly listed miners are converting power capacity from SHA-256 hashing to artificial-intelligence and high-performance computing (HPC) infrastructure.
Nine public miners generated $341 million from AI and HPC operations in H1 2026 while spending $5.11 billion on capital assets — a 15:1 capex-to-revenue ratio that reflects an industry still in early deployment. Contracted AI and HPC deal value across the sector now exceeds $70 billion. According to Bloomberg, listed miners could derive as much as 70% of their revenue from AI by end of year, up from approximately 30% at the start of 2026.
The implications extend beyond corporate strategy. A structurally lower hashrate reduces the cost of theoretical 51% attacks, raises questions about Bitcoin's long-term security budget, and introduces a new correlation between GPU-cloud demand cycles and Bitcoin mining economics.
Bitcoin's network hashrate peaked near 1,151 EH/s in late October 2025. By the end of August 2026, the smoothed seven-day average had fallen to approximately 914 EH/s, according to data tracked by CryptoTimes. This 21% decline marks the longest sustained period below an all-time high in roughly a decade.
The decline has not been linear. In late January 2026, hashrate briefly plunged to 663 EH/s — a 40% drop in days — before recovering to approximately 1,020 EH/s by March. CoinDesk reported that Q1 2026 marked the first quarterly hashrate decline in six years.
Public mining companies have shed an estimated 56 EH/s of realized hashrate during H1 2026, a 15% contraction among listed operators, according to Blockspace. The network-wide decline was approximately 10% over the same period, indicating that public miners are reducing capacity faster than private operators.
Mining difficulty has adjusted downward accordingly. Network difficulty fell 7.76% in early 2026 and dropped a further 19% cumulatively by mid-year. As of September 13, 2026, difficulty stands at 127.45T.
The post-halving economics tell the story. According to the CoinShares Q1 2026 mining report, the weighted average cost to produce one BTC among listed miners sat at roughly $80,000 in Q4 2025. Multiple sources estimate current all-in production costs between $80,000 and $90,000 per BTC, while spot price has traded in the $76,600–$78,700 range throughout September 2026. Bitcoin traded at approximately $78,737 on September 9, 2026, according to CoinDCX data.
Hashprice — the dollar revenue per petahash per second per day — reached roughly $29/PH/s in Q1 2026. As of September 13, 2026, it sits at approximately $37/PH/s/day, a modest recovery but still well below levels that make older-generation hardware viable.
Hardware efficiency thresholds have hardened. The Antminer S23 Hydro (580 TH/s, 9.5 J/TH) produces BTC at roughly $34,200 per coin at $0.07/kWh industrial rates, according to Simple Mining data from July 2026. An S21 XP runs at approximately $48,600. The legacy S19 XP costs $77,300 per coin — firmly above spot. Miners operating hardware above 15 J/TH at electricity rates above $0.08/kWh are running at a loss.
An estimated 15–20% of the global fleet is operating below breakeven, according to CoinShares estimates. This margin compression is the proximate cause of the AI pivot: the same megawatt of power capacity earns more when leased to a hyperscaler than when pointed at SHA-256.
The scale of committed AI and HPC contracts across publicly listed miners now exceeds $70 billion in aggregate. The following company-level data illustrates the scope of the transition.
Core Scientific (CORZ): Q2 2026 revenue doubled to $164.2 million from $78.6 million year-over-year. AI colocation revenue drove the increase, climbing to $136.7 million from $10.6 million — a 12.9x increase. AI colocation now accounts for 83% of quarterly revenue, up from 14% a year prior. Core Scientific announced a partnership with AMD covering up to 2.5 GW of leasable data-center capacity, anchored by 15-year agreements covering 530 MW beginning in 2027. The company recorded a $1.16 billion net loss in Q2, driven by warrant revaluation, not operations.
Cipher Digital (CIFR): Cipher holds a 15-year, 300 MW direct lease with AWS, projected to generate $5.5 billion in revenue. Its Barber Lake facility is expected to deliver Phase I (168 MW) to Fluidstack by September 2026. Q2 2026 revenue fell to $24.8 million from $43.6 million year-over-year as the company transitions from mining revenue to contracted data-center revenue. EPS came in at -$0.65.
IREN: Disclosed $2.8 billion in cloud contracts with AI developers.
Riot Platforms (RIOT): HPC revenue is projected at 13% of total for 2026, a slower transition. However, Riot contracted 241 MW at Rockdale, including 191 MW tied to a 20-year agreement worth approximately $9.1 billion. An AMD lease is expected to generate $311 million in initial revenue, with potential for up to $1 billion if extensions are exercised.
Hyperscale Data: Ceased all Bitcoin mining at its Michigan facility effective September 1, 2026. The shutdown prepares the site for AI workloads under a master services agreement with a California-based neocloud provider. The initial 20 MW contract has a 10-year term plus two five-year extensions, valued at over $1.2 billion initially, rising past $3.0 billion with an additional 32 MW option. Hyperscale Data cut its BTC holdings by 79% to fund the transition.
Marathon Digital (MARA): The largest public miner by hashrate has taken a more cautious approach, acquiring AI-relevant facilities and announcing exploratory partnerships without fully committing to the pivot. Marathon continues to hold the largest BTC treasury among listed miners.
The industry's aggregate numbers reveal a classic infrastructure-buildout profile. Nine public miners spent $5.11 billion on capital assets during H1 2026 while generating $341.2 million in directly reported AI and HPC revenue, according to Miner Weekly data. That is a 15:1 capex-to-revenue ratio.
HPC and AI revenue among comparable miners rose 52% quarter-over-quarter from Q1 to Q2 2026. For companies furthest into the transition — principally Core Scientific and Cipher — Q2 marked the first quarter in which HPC colocation or AI-cloud revenue exceeded mining revenue. The crossover point arrived.
The contract structure matters. Most deals are 10- to 20-year leases with hyperscalers, neocloud providers, or AI developers. Revenue recognition lags capital deployment by 12–24 months in typical data-center buildouts. Companies that secured contracts in 2025 are now beginning to deliver megawatts, which should narrow the capex-to-revenue gap through H2 2026 and into 2027.
Operating margins on AI colocation are reported at 80–90% once facilities are live, according to insights4vc analysis — substantially above the margin profile of Bitcoin mining at current hashprice levels.
A 21% decline in hashrate directly reduces the economic cost of a theoretical 51% attack. During the January 2026 trough, when hashrate fell to 663 EH/s, the theoretical double-spend cost fell 30–40%, according to KuCoin research.
At current levels near 914 EH/s, the 1-hour attack cost still exceeds $1.5 million in electricity alone, a figure that does not account for hardware acquisition, coordination, and opportunity costs. No chain disruptions have been observed despite the hashrate decline.
However, two structural concerns merit monitoring. First, geographic concentration: a growing share of remaining hashrate is located on U.S. power grids, making the network more vulnerable to regional disruptions such as extreme weather or policy changes. Second, the security-budget question: if miners continue diverting power away from Bitcoin, the network's long-term reliance on transaction fees to fund security — a concern debated since Bitcoin's inception — becomes more acute as block subsidies continue halving.
Mining difficulty adjustments have partially compensated. The 19% cumulative difficulty reduction through mid-2026 has lowered the bar for remaining miners, stabilizing hashrate at current levels. The difficulty mechanism is functioning as designed, but it cannot create economic incentive where none exists.
Mining equities have decoupled from Bitcoin's spot price in 2026. The AI pivot narrative has repriced several stocks on data-center multiples rather than BTC production metrics.
As of September 10, 2026, CIFR traded at $15.92 (up 9.4% YTD despite Q2 revenue decline), MARA at $11.43, and RIOT at $21.20, per 24/7 Wall Street data. All three fell 4–6% on September 10 on a broad risk-off tape, but the year-to-date pattern shows AI-forward miners outperforming pure-play Bitcoin miners.
In May 2026, The Block reported that Cipher and Hut 8 hit fresh highs as the "miner-to-AI boom" accelerated. IREN traded at $46.27 in January 2026 — a valuation driven entirely by its $2.8 billion AI contract book rather than mining output.
The market is pricing these companies on contracted AI revenue and megawatt-delivery timelines. Companies that have announced large AI deals but have not yet delivered operational capacity trade at a premium to their current earnings, creating execution risk.
The Bitcoin mining industry is undergoing a structural transformation from proof-of-work computation to general-purpose data-center operations. The economic logic is straightforward: at current BTC prices and post-halving block rewards, the same megawatt of power capacity generates higher and more predictable returns when leased to AI workloads than when directed at SHA-256 hashing.
The transition is in its early stages. The 15:1 capex-to-revenue ratio indicates that the industry is spending heavily on infrastructure that has not yet reached commercial operation. Contract terms of 10–20 years with hyperscalers provide long-dated revenue visibility, but execution risk is material: construction delays, permitting issues, and power-grid interconnection timelines have historically challenged data-center buildouts.
For Bitcoin's network, the implications are measurable but not yet critical. Hashrate has declined 21% and difficulty has adjusted downward accordingly. The network continues to produce blocks on schedule. The deeper question — whether Bitcoin's security budget can sustain adequate hashrate as block subsidies diminish and miners diversify revenue — remains unresolved. The AI pivot has made that question more urgent.