Bitcoin mining electricity demand reached an annualised 190 terawatt-hours (TWh) in December 2025, up 38% from 138 TWh recorded in June 2024, according to preliminary findings from the Cambridge Centre for Alternative Finance (CCAF). Alexander Neumueller of the CCAF presented the figures at the Energy Investors Forum in Dallas.
The 52 TWh increase came alongside a shift in the energy mix: hydropower has overtaken natural gas as Bitcoin mining’s single largest power source, and low-carbon supply now accounts for 59.4% of the reported mix, up from 52.4% in the prior study.
Bitcoin Mining Electricity Demand and the Hydro Shift
The first edition of the CCAF Digital Mining Industry Report, published in April 2025, surveyed 49 digital mining firms across 16 jurisdictions with operations in 23 countries, capturing roughly 48% of global Bitcoin hashrate. At that point, natural gas supplied 38.2% of miners’ electricity, making it the largest single source. Renewables collectively contributed 42.6%, nuclear added 9.8%, and coal had fallen to 8.9% from 36.6% in the 2022 estimate.
The preliminary update reverses the top slot. Hydropower now leads, with Cambridge pointing to stronger survey coverage in hydro-rich markets, particularly Ethiopia, where miners have scaled up around cheap generation from the Grand Ethiopian Renaissance Dam. The full source-by-source breakdown will follow in the second edition, expected later in 2026.
The first report also found that annual electricity consumption had grown 17% year-on-year to 138 TWh, representing around 0.54% of global electricity usage, even as hardware efficiency gains partially offset the impact of rising hashrate. The new preliminary data shows that dynamic accelerating: more machines, more hashrate, and efficiency gains that still cannot keep pace with network growth.
Greener Mix, But Emissions Still Rose 20%
A cleaner supply mix slowed emissions growth but did not stop it. Cambridge’s estimate for Bitcoin mining greenhouse-gas output rose from roughly 40 million to 48 million tonnes of CO₂ equivalent, a 20% increase, even as the low-carbon share of reported supply climbed.
The methodology gap in Cambridge’s own numbers is worth keeping in mind. The survey-based approach, which covers slightly more than half of global hashrate, produced an estimate of 39.8 million tonnes in the prior edition. A separate location-based model from the same institution produced 69.6 million tonnes for the same period. Results depend heavily on assumptions about mining locations, grid mixes, and the treatment of stranded or flared energy. The Cambridge Bitcoin Electricity Consumption Index (CBECI), a distinct daily index model rather than a survey, estimated annualised Bitcoin network consumption at roughly 175 TWh as of early 2025, using different inputs from the survey-based report.
Cambridge also flagged that U.S. companies supplied a disproportionate share of survey responses, which likely overstates the country’s share of global mining activity. The rise in reported hydropower may partly reflect better coverage of Ethiopian and other hydro-dependent operators rather than a pure shift in the underlying energy mix.
AI Deployments: Intent Is Not Commitment
The survey examined how many miners have moved capacity into artificial intelligence and high-performance computing (HPC). Only 10% of respondents had already allocated power to AI or accelerated computing services. More than 40% of the remainder said they were exploring the option.
Neumueller was direct on that gap: ‘intent to look into it is not commitment to deploy.’ AI data centres require costly networking, precision cooling, and reliability guarantees that typical Bitcoin mining sites are not built to provide. Bitcoin loads can be curtailed when power prices spike; AI customers want steady supply and contractual uptime. A Congressional Research Service report noted that large flexible loads such as crypto mining could represent 10% of total Texas grid consumption in 2025, which illustrates why grid operators and AI hyperscalers are watching the same sites.
Despite the operational hurdles, nearly nine in ten survey respondents expected AI and HPC diversification to gain ground over the next several years, and listed miners have announced more than $70 billion in combined AI and HPC contracts.
TeraWulf’s Q1 2026 earnings show what that transition can look like in practice. The company reported $21.022 million in HPC lease revenue for the quarter against $12.990 million in digital asset mining revenue, with 60 MW of operational critical IT HPC capacity for Core42 already live at its Lake Mariner facility. Total Q1 2026 revenue came to $34.012 million, compared with $34.405 million from digital asset mining alone in Q1 2025. The revenue base is similar in size; the composition has inverted.
Additional TeraWulf preliminary Q1 2026 results confirmed the company was nearing completion of its CB-3 building, with CB-4 and CB-5also scheduled for delivery in 2026, pointing to further HPC capacity coming online across the year.
Cambridge’s full second-edition report will deliver a finalised energy breakdown and updated emissions methodology. Given the divergence between the survey-based and location-based estimates, the methodological notes may matter as much as the headline figures.
