Loading article...
What makes a blockchain sustainable, how much energy different chains actually use, and how to choose or run a low-impact network. Verified data from ethereum.org, CCRI, and Cambridge CCAF.
A sustainable blockchain secures transactions with minimal energy, hardware waste, and carbon per unit of useful work. It does this by choosing an efficient consensus mechanism, keeping node requirements light, and accounting for where its electricity comes from.
Sustainability is not a marketing label. It is measurable: annual electricity in kilowatt-hours, carbon intensity in grams of CO2e per kilowatt-hour, total emissions in tonnes CO2e per year, and the share of renewable energy powering validators.
This guide explains what drives those numbers, how major chains compare with verified sources, and how you can evaluate or reduce footprint if you build, operate, or invest in Web3.
If you want the short answer: the biggest drop in blockchain energy use came from replacing proof-of-work with proof-of-stake. Everything else is optimization around that decision.
Three inputs determine a network's footprint. Each can be audited.
In proof-of-work, security comes from miners burning electricity and hardware to solve puzzles. In proof-of-stake, security comes from validators locking capital that can be destroyed if they misbehave. The second approach needs no race for hashes.
Total consumption equals average power per node times number of nodes. A network with 300 low-power nodes can use less per year than a network with 1,000 high-power nodes, even if the second does far more transactions.
Two networks with the same kilowatt-hours can have different emissions if one runs on coal-heavy grids and the other on hydro or wind. CCRI applies country and, for the United States, state-level emission factors to each node's location. Cambridge CCAF tracks the share of sustainable energy separately.
Per-transaction energy is often quoted, but treat it carefully. As ethereum.org notes, the energy to propose and validate a block is independent of how many transactions fill that block. Layer-2 rollups make this even more misleading, because a single layer-1 batch can settle thousands of layer-2 transactions with little extra energy. Always check both per-node and annual totals.
Ethereum is the clearest before-and-after case. Before the Merge on September 15, 2022, Ethereum ran proof-of-work. According to CCRI via ethereum.org, it used about 22,900,320 MWh per year and emitted about 11,016,000 tonnes CO2e. After the Merge, CCRI measured 2,601 MWh per year and 870 tonnes CO2e, using regional carbon intensities across measured client configurations covering 95.54 percent of clients. That is a reduction of 99.988 percent in electricity and 99.992 percent in carbon. The Cambridge Blockchain Network Sustainability Index provides a rolling 7-day average with a slightly different method; an updated Cambridge assessment cited by Cambridge Judge Business School in 2025 put the post-Merge figure at 7.87 GWh per year and 2.37 ktCO2e, still in the same low-thousands Megawatt-hour range. For context, ethereum.org compares post-merge Ethereum at 0.0026 TWh per year to Bitcoin at 149 TWh, global data centers at 190 TWh, PayPal at 0.26 TWh, and an average United States household at about 10,600 kWh per year.
The mechanism is specific. Time is divided into 12-second slots, 32 slots per 6.4-minute epoch. A validator chosen by RANDAO proposes a block, committees attest, and Casper FFG finalizes checkpoints when two-thirds of staked ETH vote correctly. A Raspberry Pi-class machine can run a node per ethereum.org documentation, not an ASIC farm.
CCRI systematically measured six proof-of-stake networks in January 2022 with the same bottom-up method. The table below uses that snapshot so numbers are comparable. All figures change as nodes join or leave and as throughput changes.
| Network | Nodes | Tx per year | Electricity per node (kWh/yr) | Electricity per tx (Wh) | Total electricity (kWh/yr) | Carbon (tCO2e/yr) |
|---|---|---|---|---|---|---|
| Cardano | 3,002 | 11.9m | 199.45 | 51.59 | 598,755 | 284.41 |
| Polkadot | 297 | 4.0m | 236.49 | 17.42 | 70,237 | 33.36 |
| Solana | 1,015 | 11.8b | 1,938.85 | 0.166 | 1,967,930 | 934.77 |
| Tezos | 375 | 2.5m | 302.00 | 41.45 | 113,249 | 53.79 |
| Avalanche | 1,084 | 93.9m | 451.39 | 4.76 | 489,311 | 232.42 |
| Algorand | 1,190 | 190.0m | 430.82 | 2.70 | 512,671 | 245.52 |
Source: CCRI, Energy efficiency and carbon emissions of PoS Networks, Jan 2022.
What the table shows:
Later CCRI reports add more chains with the same method:
Not in the CCRI six but relevant: Hedera Hashgraph does not use blocks in the same way, it uses a directed acyclic graph with gossip, and its council structure yields similar low per-node draw claims. IOTA's Tangle is also graph-based and reports among the lowest per-transaction values when idle, but both should be checked against annual totals and node counts from the same methodology rather than headline per-transaction claims.
Layer-2 rollups are the most direct way to cut energy per useful transaction without changing layer-1. Transactions execute off-chain, a sequencer posts a batch to layer-1, and the batch cost is shared across hundreds or thousands of transactions. The sequencer adds a small server's draw, the number of layer-1 blocks does not scale linearly with transactions, and per-transaction energy falls by roughly two orders of magnitude versus counting base-layer transactions only. This is why ethereum.org warns that comparing chains on per-transaction alone can be gamed, and why Polygon's accounting now separates its own draw from its allocated share of Ethereum.
Measurement matters. CCRI's approach since 2022 is to measure actual hardware draws for each client on several machines, weight by observed node distribution, and multiply by grid intensity for node locations. Cambridge CCAF's Bitcoin index uses a hybrid top-down model assuming miners use only profitable hardware, then converts to an annualized TWh via a 7-day average. Both groups flag limits: location data is incomplete, hardware mix shifts, and hashrate volatility adds noise. That honesty is why both provide lower, best guess, and upper bounds. For Ethereum post-Merge, CCRI gave 278.60 lower, 869.78 best guess, and 2,090.32 upper tonnes CO2e per year as of August-September 2022.
For Bitcoin, the April 2025 Cambridge Digital Mining Industry Report surveyed 48 percent of global mining activity and found:
Using that range, post-merge Ethereum at 0.0026 TWh is about 53,000 to 57,000 times lower than Bitcoin in the same tables.
Offsets are the last step, not the first. Algorand has been carbon neutral since 2021 and blocks its offset purchases on-chain: each epoch's footprint is notarized by a sustainability oracle and an equal amount of carbon credit is locked as an Algorand Standard Asset into a green treasury. The Foundation lists the actual transactions: 2022 Evergreen REDD+ in Brazil, 2023 UNITOR REDD+ in Brazil, 2024 Phlogiston Phase I thermal reduction in the United States, 2025 Energy from Renewables in Maharashtra wind in India. Projects include Southern Cardamom REDD+ (over 3 million tonnes avoided per year), Vichada Gold Standard reforestation in Colombia, Oaxaca Wind covering 700,000 Mexican homes, and Sumatra Merang Peatland restoration. These are verified credits from ClimateTrade's marketplace, not self-reported estimates.
What proof-of-stake improves:- Energy drops by more than 99 percent for the same security budget, because security no longer requires continuous hashing. Ethereum's documented 99.988 percent fall is the largest verified example.
What it does not fix automatically:- Geographic concentration matters. If many validators sit in coal-heavy grids, emissions rise even though kilowatt-hours do not. Location data in CCRI's studies was missing for 1.95 to 4.1 percent of nodes, for which world average intensity was assumed.
If you are evaluating chains: 1. Start with annual electricity and carbon, not per-transaction. Use CCRI's API or reports and Cambridge CCAF's indices for Bitcoin and Ethereum. Both are updated methods, not one-time press releases.
Check node hardware specs and count. A chain that requires 128 GB RAM and 2 TB NVMe per validator will draw more per node than one that runs on 8 GB and 100 GB SSD, even before transactions.
Ask for energy mix. Cambridge now reports 52.4 percent sustainable for Bitcoin mining; Algorand and Ethereum reports report 56.4 percent renewable share in recent Ethereum modeling and project-specific mixes for offset portfolios. If a team cannot tell you how they map nodes to grid factors, treat the carbon claim as incomplete.
For EU reporting, request MiCA sustainability indicators. CCRI publishes MiCA-compliant fact sheets for Cardano and others with the exact disclosures ESMA proposes.
If you are running infrastructure:
On Ethereum, a single validator needs 32 ETH (or up to 2,048 ETH with compounding 0x02 credentials after Pectra on May 7, 2025), plus an execution client, a consensus client, and a validator client. Power draw for a home setup with current clients is around a few tens to about 100 watts continuous, roughly one modern desktop, not a mining rack. EthStaker and ethereum.org both put a full home staker at about 100 watts. Keep the machine on a low-carbon grid if you can; the location changes your attributed emissions more than the wattage does.
If you hold less than 32 ETH, liquid staking pools or pooled operators let you stake from 0.01 ETH with a bond of about 1.5 to 4 ETH in some designs. You avoid running hardware but add middleware risk. Compare operator diversity and whether they run distributed validator technology across geographies and clients.
Run only the clients you need, keep storage at the recommended prune level, and update clients to lower-draw versions measured by CCRI. Do not run proof-of-work miners alongside proof-of-stake validators on the same power without accounting for the mining draw separately. If you are building apps:- Deploy on proof-of-stake layer-1 and push high-volume actions to a rollup or validium. Batching alone can cut attributed energy per transaction by more than 10 times even before consensus savings.
Yes on energy. Every like-for-like measurement since CCRI's 2022 studies shows proof-of-stake networks in the hundreds of megawatt-hours per year, while proof-of-work Bitcoin measures in the hundreds of terawatt-hours per year. The gap is 4 to 5 orders of magnitude. Carbon depends also on grid mix, but even with coal-heavy hosting, the smaller kilowatt-hour base keeps proof-of-stake lower.
Because energy per transaction divides a fixed per-node cost by transaction count. Solana's per-node draw is the highest at 1,938.85 kWh per year due to higher hardware needs, but it divides that across 11.8 billion transactions per year, giving 0.166 Wh per transaction. Polkadot divides a smaller 70,237 kWh total across 4.0 million transactions, giving 17.42 Wh per transaction but a much smaller annual footprint.
Algorand's mechanism is Pure Proof-of-Stake, which is among the lowest draw per the Algorand energetic model at 265 tCO2 per year as of June 2024. The chain is made carbon negative by the Algorand Foundation purchasing verified credits on ClimateTrade and locking them on-chain in a green treasury. The chain itself is low emission; negativity comes from offsets that are recorded on-chain with block IDs for audit, not from zero energy.
No. Gas fees depend on block space demand versus capacity, which the Merge did not change. Block time went from a variable average of about 13.3 seconds to a consistent 12 seconds, about 10 percent more frequent, but not a user-noticeable speed change. Finality changed from probabilistic to explicit.
Proof-of-work waste includes ASIC turnover and cooling water. Proof-of-stake waste is standard server e-waste and data-center cooling, far smaller in volume. Cambridge and CCRI focus on electricity and carbon; water and mining hardware lifecycle are tracked separately in emerging models but directionally fall with proof-of-stake because specialized hardware is not required.
MiCA requires issuers to publish sustainability indicators including electricity consumption and carbon footprint using a prescribed method (ESMA). CCRI's methodology papers from December 2024 and June 2024 document how to compute these, and fact sheets already exist for Cardano, CELO, Polymesh, Polygon, TRON, and Ethereum. If you operate in the EU, use those definitions rather than internal estimates.
If I already use Bitcoin, how can I lower attributed impact? Use layer-2 like Lightning for frequent payments, choose miners or custodians that report energy mix with location proof, and separate accounting for mining versus transaction services. Cambridge's finding that natural gas now exceeds coal and that hydro and wind are the largest renewable components suggests mix is shifting, but the absolute draw remains high at 138 to 183 TWh per year.
A blockchain is sustainable when it secures the network with locked capital instead of burned electricity, runs on light hardware, and can show where its power comes from with a method you can replicate. Ethereum's move to proof-of-stake cut its electricity by 99.988 percent to 2,601 MWh per year and its carbon to 870 tonnes CO2e per year per CCRI on ethereum.org. Among proof-of-stake peers, annual totals range from Polkadot at about 70 MWh to Solana at about 1,968 MWh in the same comparable snapshot, with TRON at 163 MWh and Polygon at about 119 MWh including its Ethereum allocation. Bitcoin remains at 138 to 183 TWh per year even as its sustainable share rose to 52.4 percent in Cambridge's 2025 survey.
Choose on annual totals and verifiable mix, use layer-2 for scale, and prefer chains that publish client-level measurements and retire credits on a public registry. That is the practical meaning of "green" in Web3.
Explore more guides and career playbooks