The Cambridge Centre for Alternative Finance has estimated Ethereum's annual electricity consumption at 7.87 gigawatt-hours following the Merge, a reduction of over 99.9% from its pre-transition level of about 2.4 gigawatts continuous power demand. The study, based on direct node measurements, places Ethereum near the bottom of energy intensity among major proof-of-stake blockchains when adjusted for market value, consuming roughly 33 kWh per $1 million of market cap.
Post-Merge energy profile
Cambridge researchers measured 8,522 discoverable full nodes and calculated a network-weighted average power draw of 105 watts per node. They tested 20 combinations of execution and consensus clients on two hardware setups: a residential system drew a median of 18 watts, while a professional workstation drew about 152 watts. About 36% of nodes ran on residential hardware, with the remaining 64% hosted in cloud or enterprise data centers. The United States hosted 31% of nodes, followed by Germany at 16%, Finland at 8%, and France at 6%.
Ethereum's annual emissions now stand at roughly 2.37 kilotonnes of CO2 equivalent, driven primarily by the grid mix powering its nodes. Renewable energy supplied 39.4% of the network's power, nuclear 17%, and fossil fuels 43.6% (natural gas the largest at 27.7%). Alexander Neumüller, research lead at Cambridge's digital assets energy program, noted that under proof-of-stake, electricity is no longer the price of security.
Comparison with other PoS networks
Among major proof-of-stake blockchains, Ethereum's absolute consumption of 7.87 GWh is second only to Solana's 13.48 GWh. However, when normalized by market value, Ethereum ranks second-lowest, behind BNB Chain, at 33 kWh per $1 million. Solana consumed about 283 kWh per $1 million, roughly 8.5 times higher. The top-tier PoS networks studied—including NEAR, Tron, TON, Cardano, and BNB Chain—used 38 GWh combined. Cambridge avoided per-transaction estimates because about 92% of Ethereum ecosystem transactions now settle on scaling networks, making mainnet-only calculations incomplete. The report treats future demand as uncertain, noting that lighter stateless verification could lower hardware requirements but wider node participation could offset those savings.