Bitcoin's annualized electricity consumption has fallen from an estimated 190 TWh in December 2025 to approximately 138 TWh by mid-2026, according to Cambridge Centre for Alternative Finance data. The network hashrate, which peaked near 1,160 EH/s in October 2025, has declined to roughly 900 EH/s...
"Mining expands only while Bitcoin's expected real price growth exceeds roughly 18.92% a year, the rate needed to offset the 50% cut in new coins every four years." — Saifedean Ammous, Author, The Bitcoin Standard
Bitcoin's annualized electricity consumption has fallen from an estimated 190 TWh in December 2025 to approximately 138 TWh by mid-2026, according to Cambridge Centre for Alternative Finance data. The network hashrate, which peaked near 1,160 EH/s in October 2025, has declined to roughly 900 EH/s as of mid-September 2026—an 18.3% drawdown from peak. Hashprice, the standard measure of per-unit mining revenue, closed Q1 2026 at $23.9 per PH/s per day, the lowest reading since 2018.
These three data points converge on a structural thesis that is gaining traction among researchers: Bitcoin mining may have already reached its peak electricity draw. A September 15, 2026 essay by economist Saifedean Ammous formalized the argument, framing it as a testable hypothesis tied to halving mathematics, declining price appreciation rates, and competition from AI data center demand. Meanwhile, a CoinShares Q2 2026 report published on September 15 documented that at least 35 EH/s of hashrate from publicly listed firms is exiting the network, calling the pivot to AI infrastructure "structurally irreversible." The convergence of post-halving economics, hardware migration, and sovereign resource constraints is reshaping the mining industry's energy footprint in real time.
The Cambridge Bitcoin Electricity Consumption Index (CBECI) provides the most widely cited estimate of Bitcoin's power draw. Following a methodology revision in 2023 that corrected for periodic overestimation, the index placed Bitcoin's annualized electricity consumption at approximately 190 TWh in December 2025. By mid-2026, the central estimate had declined to roughly 138 TWh—equivalent to approximately 0.5% of global electricity production.
The decline tracks directly with hashrate. The Bitcoin network reached approximately 1,160 EH/s in October 2025. By July 31, 2026, the figure had dropped to 878 EH/s, according to CoinWarz data. As of mid-September 2026, the network sits at approximately 900 EH/s, with difficulty at 127.45T—18.3% below its peak.
The correlation between hashrate and electricity consumption is near-linear for a given hardware generation. Fewer machines hashing means fewer watts consumed. The decline is not a temporary dip tied to seasonal factors; it reflects a sustained drawdown in deployed mining capacity.
The April 2024 halving reduced Bitcoin's block subsidy from 6.25 BTC to 3.125 BTC, cutting miners' guaranteed per-block revenue by 50% overnight. At current prices near $80,000, the network issues approximately 450 BTC per day, generating roughly $36 million in daily subsidy revenue across all miners.
Ammous's September 2026 analysis centers on a specific mathematical threshold: Bitcoin's price must compound at approximately 18.92% annually in real terms to offset each halving's 50% reduction in new coin issuance over the four-year cycle. When actual price appreciation falls below this hurdle rate, marginal miners reduce investment and power consumption. The thesis does not claim mining will contract to zero—only that the rate of expansion has structurally decelerated.
The data supports this framing. Bitcoin traded at approximately $63,000 at the time of the April 2024 halving and sits near $80,000 in mid-September 2026—a roughly 27% gain over 29 months, annualizing to approximately 11%. That figure falls below the 18.92% threshold Ammous identified, suggesting that the current price trajectory is insufficient to incentivize network-wide hashrate expansion.
CoinShares' Q2 2026 mining report found that the average cash cost to produce one bitcoin reached approximately $75,500 in Q2 2026, while bitcoin ended the quarter at $58,400. By September, prices have recovered above $80,000, restoring some margin—but the all-in production cost for many operators exceeds $88,000, according to analyst estimates. The industry is operating near its cost of production, not well above it.
CoinShares researcher Luke Nolan's Q2 2026 report documented what he termed a "structurally irreversible" migration of publicly listed mining companies toward AI and high-performance computing (HPC) infrastructure. The economics are stark: AI infrastructure generates approximately $1.5 million in revenue per megawatt of capacity, compared to $0.5 million for Bitcoin mining—a 3:1 ratio favoring compute over hashing.
Specific company data illustrates the scale of the shift:
Core Scientific (CORZ): Expanded its CoreWeave contract to $10.2 billion over 12 years. AI and HPC colocation reached 39% of Q4 revenue. The company paid $41.9 million to terminate a mining hardware agreement, cancelling roughly 15 EH/s of ASIC deliveries. LTM revenue reached $510.67 million, with approximately 350 MW energized for AI workloads.
IREN: Total revenue reached $707 million for the fiscal year ending June 30, 2026, up from $501 million a year earlier. HPC revenue is projected to reach 71% of total revenue, up from 3% in 2024. The company plans to complete its exit from Bitcoin mining by December 31, 2026.
Keel (formerly Bitfarms): Completely halted mining operations in June 2026.
Cipher Digital (CIFR): Expected to withdraw from mining by end of 2027.
In aggregate, CoinShares estimates at least 35 EH/s of hashrate from publicly listed firms is set to exit the Bitcoin network—representing 4.7% of the approximately 750 EH/s contributed by listed miners. Many operators have signed AI colocation leases extending 15 years or longer, eliminating optionality to return to mining even if Bitcoin prices surge. As CoinShares noted, these operators "won't be coming back."
The structural logic is straightforward: mining companies hold power purchase agreements, land, cooling infrastructure, and grid interconnections. AI data center demand—driven by persistent GPU shortages flagged by Nvidia across multiple quarters—provides a higher-margin use for those same assets. The conversion is one-directional because it requires physical hardware replacement and long-term contractual commitments.
Ethiopia provides a real-time case study in the fragility of mining's energy access assumptions. On September 15, 2026, Bloomberg reported that Ethiopian Electric Power (EEP) had slashed electricity deliveries to Bitcoin miners to 23% of contracted levels, down from earlier step-downs at 75% and 50%.
The cause is hydrological: El Niño conditions reduced water inflows into Ethiopia's reservoirs by 20%. Hydropower accounts for approximately 95% of EEP's generation mix, making electricity production acutely sensitive to rainfall variation. Bitcoin miners had become EEP's largest revenue-generating customer segment, accounting for 35% of the utility's revenue and consuming nearly one-third of the nation's total 9,730 MW production capacity.
EEP CEO Ashebir Balcha stated the company cut power to prioritize households and manufacturers. The utility plans to reassess conditions in October and may impose further reductions or restrict electricity exports to neighboring countries.
The Ethiopia case undermines a persistent narrative in mining advocacy: that miners primarily consume "stranded" or "surplus" energy that would otherwise go unused. When sovereign resource constraints bind, mining is the first load to be curtailed. The same dynamic has played out in Kazakhstan, Iran, and regions of Texas during grid stress events. Mining's interruptibility is simultaneously its selling point as a grid-balancing tool and its vulnerability as a business model dependent on continuous uptime.
U.S. policymakers are responding to the shifting mining landscape with legislation that attempts to both support and regulate the industry. The Mined in America Act (S. 4251), introduced by Senators Cassidy and Lummis on March 30, 2026, combines three elements: a voluntary Department of Commerce certification program for mining facilities, domestic hardware manufacturing incentives, and a codification of the Strategic Bitcoin Reserve into law.
The bill addresses a hardware dependency highlighted by an April 2026 Congressional Research Service report (R48914): while the United States controls an estimated 38% of global Bitcoin hashrate, approximately 97% of specialized mining ASICs are manufactured by Chinese firms, primarily Bitmain and MicroBT. The bill would require certified facilities to phase out hardware from foreign adversaries by January 1, 2030, and would allow certified miners to sell newly mined bitcoin directly to the government with a capital gains tax exemption.
The CRS report also documented that 11 cryptocurrency mining operators in New York represent approximately 7.7 TWh of electric demand, with operators potentially adding 790 MW of increased load through 2026. The Senate Select Committee on Intelligence flagged concerns about Chinese-owned mining facilities operating on U.S. soil.
With Bitcoin trading near $80,000 as of September 19, 2026, the mining industry's margin picture is bifurcated:
Hashprice—the revenue a miner earns per unit of hashrate—sat at approximately $29/PH/s/day in July 2026, recovering from the Q1 2026 low of $23.9. For context, hashprice was approximately $70/PH/s/day during the COVID-era recovery in 2020. A recent September adjustment pushed difficulty up 1.31%, with hashprice rising 22% on an improved BTC price. But the long-term compression trend remains intact: each halving structurally reduces the revenue per hash absent proportional price increases.
Electricity consumption has declined 27% from peak. Cambridge CBECI data shows Bitcoin's annualized draw falling from ~190 TWh (December 2025) to ~138 TWh (mid-2026). Hashrate has dropped 22% from its October 2025 peak.
The AI pivot is a one-way door. At least 35 EH/s from listed miners is exiting the network. Multi-year AI contracts and physical hardware conversion make reversion to mining economically irrational. Revenue per MW from AI is 3x that of mining.
Halving math constrains expansion. Bitcoin's price must compound at ~19% annually to offset each halving's subsidy reduction. Current annualized appreciation since the April 2024 halving is approximately 11%, below the required threshold.
Sovereign curtailment is real and growing. Ethiopia's 77% power reduction to miners demonstrates that "stranded energy" narratives collapse under genuine resource scarcity. Mining is consistently the first industrial load curtailed during grid stress.
U.S. policy is bifurcated. The Mined in America Act seeks to reshore mining and codify the Strategic Bitcoin Reserve, while CRS reports document growing Congressional concern about mining's grid impact and foreign hardware dependency.
The thesis that Bitcoin mining electricity consumption has peaked is no longer speculative—it is an observable trend supported by Cambridge consumption data, hashrate metrics, and the documented exit of major mining operators from the network. The structural drivers are mutually reinforcing: halving economics compress margins, AI demand offers a higher-value use of the same infrastructure, and sovereign resource pressures periodically remove mining capacity from grids.
None of this implies Bitcoin's network is at risk. Difficulty adjustments ensure the network continues to function regardless of hashrate level. The security implications of declining hashrate are a separate question, but the network has operated securely at far lower hashrate levels. What has changed is the growth trajectory: the era in which Bitcoin mining's electricity consumption expanded monotonically with each cycle appears to have ended.
Whether the current ~138 TWh level represents a permanent ceiling or a cyclical trough depends on a single variable: Bitcoin's price trajectory relative to the 18.92% annual hurdle rate. If prices compound above that threshold through the next halving in 2028, mining will expand again. If they do not, December 2025 may prove to have been the high-water mark.