Tezos (XTZ) sustainability report

Tezos is present on the following networks: Tezos.

Tezos runs a liquid proof-of-stake network on top of a classical Byzantine fault-tolerant agreement algorithm called Tenderbake, adapted from the Tendermint family. Validators, known on this network as bakers, are allocated the right to propose blocks and to attest them in proportion to their baking power, which is computed once per cycle from the funds they have frozen themselves, the funds other holders have frozen alongside them, and the balances merely delegated to them, with frozen funds weighted more heavily than delegated ones. Within a block's round a designated proposer publishes a candidate and a committee of attesters votes on it; agreement requires more than two thirds of the committee's weight. Finality is deterministic rather than probabilistic, arriving within a couple of blocks, and block intervals were shortened to six seconds by the amendment adopted in January 2026.

The liquid part of the design is delegation that does not move custody. A holder can point the weight of a balance at a baker without transferring the coins, without freezing them and without surrendering the ability to spend them; the baker gains voting weight but never control. A distinct and stronger commitment, staking, freezes funds alongside the baker's own and shares the baker's exposure to penalties, while plain delegation carries no such exposure. Bakers may accept external frozen stake up to a multiple of their own.

What most distinguishes the network is that it amends itself on chain. A proposed protocol change moves through five automatically scheduled voting periods, and if it carries it activates at the end of the last one without a hard fork, coordinated release or chain split; the code the network runs is replaced by the network itself. More than twenty amendments have taken this route. Recent ones reshaped consensus and its economics: the January 2025 amendment bounded issuance against a target staked ratio and raised the external stake limit, the May 2025 amendment shortened cycles to roughly a day and tied part of baker reward to data availability participation, the September 2025 amendment introduced aggregated attestation signatures, and the amendment active since June 2026 widened data availability bandwidth and made attestation lag dynamic.

Tezos is present on the following networks: Tezos.

Rewards on Tezos are paid for proposing blocks, for attesting them and for participating in the data availability layer, with a share of baker reward conditional on that last duty. Issuance is adaptive rather than fixed: the rate at which new units of the native asset are created moves within bounds according to how much of the supply is frozen, rising when the frozen proportion falls below a target of roughly half and falling as it rises above, so the protocol leans against both under- and over-participation instead of paying a constant rate. Three roles share the proceeds. A baker runs the infrastructure and takes a fee on what it distributes. A staker freezes funds alongside a baker, earns the higher rate and accepts the baker's penalty exposure. A delegator lends only weight, keeps its funds liquid, earns less and bears no penalty exposure at all.

Penalties are real and are applied to frozen funds. Signing two different blocks at the same level and round, or attesting two conflicting proposals, is denounced by evidence included on chain, and the offending baker's frozen funds, together with the frozen funds of those staking with it, are cut at the end of the cycle. Double baking draws a fixed proportion. Double attesting draws an adaptive proportion that grows with the square of the share of the committee that equivocated together, so an isolated accident costs little while a coordinated attack can consume the entire frozen balance. A fraction of what is taken rewards whoever included the evidence and the remainder is destroyed, and a denounced baker is barred from baking and attesting for a period. A baker that simply goes quiet is deactivated rather than penalized, and must reactivate to resume.

Users pay a transaction fee that scales with the size of the operation in bytes and the gas it consumes, and that fee goes to the block proposer. Storage is charged separately and permanently: allocating space in the ledger, originating a contract or creating a new account destroys an amount proportional to the bytes claimed rather than paying it to anyone. Smart contract execution is metered in gas against per-operation and per-block ceilings, and rollup and data availability operations carry their own costs.

Tezos is present on the following networks: Tezos.

Because this is a stake-based network with no computational race, the energy estimate is constructed from the machines that keep it running rather than from any model of mining economics. The relevant population is more varied than a single validator count suggests: bakers and the attesting infrastructure behind them, the accompanying nodes that serve and sample the data availability layer, the nodes operating smart rollups, and the ordinary full nodes that relay and validate without taking part in consensus. Their number is estimated from publicly observable network data, from peer crawling, and from on-chain records of which addresses hold baking rights, which makes the consensus-active portion of the population unusually well bounded even though the servers behind those addresses are not directly enumerable.

Hardware is inferred rather than surveyed. The published operating requirements for the node software, covering processor, memory, disk and bandwidth, indicate what a competent operator would deploy, and the electrical draw of machines of that class is taken from controlled bench measurement at load and at rest rather than from nameplate figures. The annual total is the aggregate across the estimated population including idle draw, since a baking node consumes power continuously regardless of how often it is called on. Where a token issued on this network is being reported rather than the network itself, a portion of the network total is attributed to it using observed on-chain transfer volumes.

Two caveats are specific to this network. Its capabilities are extended by on-chain amendment, and successive amendments have changed cycle length, block interval, attestation duties and data availability bandwidth; each of those alters what a node must store, verify and transmit, so a hardware profile inferred before an amendment can understate requirements after it. Separately, a baker may run several machines for redundancy and signing separation, which a public count of endpoints will not reveal. More broadly, the population and hardware mix are estimates from public observation, not metered readings; where evidence is thin the assumption chosen is the one more likely to overstate consumption than understate it; and figures are restated as observation improves.

Tezos is present on the following networks: Tezos.

The renewable share follows from where the hardware sits, so the first task is placing it. Locations are inferred from publicly observable network information: the addresses nodes advertise to peers, the autonomous systems and hosting providers those addresses belong to, and what operators publish about themselves. Many bakers are publicly identified services that state where they operate, which helps, but the address of a node still identifies a facility rather than an owner, and it is the facility that draws current from a grid, so hosting location takes precedence over any nationality attributed to the operator.

The picture is inevitably partial. Signing infrastructure is commonly kept off the public network behind a remote signer, redundant machines are not separately visible, and nodes that accept no inbound connections cannot be observed at all. Where the geographic spread cannot be pinned down from the network's own data, the distribution observed on a network with a comparable operating profile stands in for it, selected for similar participation requirements, similar operator incentives and similar hosting behavior. That substitution is the largest single source of uncertainty in the renewable figure and a reason it is less firm than the consumption estimate beneath it.

Locations are then weighted by the consumption attributed to them and matched against regional electricity statistics, so each part of the network's draw inherits the generation mix of its supplying grid. The renewable proportion reported is that consumption-weighted average, which is not the same as the proportion of nodes situated in countries with clean grids. The underlying statistics are annual averages and carry no hourly resolution, so time-of-day variation in the mix is invisible to the method.

Energy intensity is expressed at the margin: the extra electricity associated with one additional transaction at the network's current node population and throughput. Since these nodes run continuously and their draw hardly varies with how busy the chain is, that marginal quantity is small and shrinks as throughput grows, a consequence of how the measure is defined rather than a change in the machines. Regional generation mix is drawn from Share of electricity generated by renewables, compiled by Our World in Data from Ember and the Energy Institute's Statistical Review of World Energy.

Tezos is present on the following networks: Tezos.

Emissions figures are calculated, not metered. Each geographically attributed slice of the network's estimated electricity use is multiplied by the carbon intensity of the grid serving that location, using the same placement of nodes that supports the generation-mix analysis: advertised addresses and hosting registrations locate consumption, a structurally comparable network fills the gaps where placement cannot be established, and the result is a consumption-weighted distribution across grids rather than a tally of nodes by country.

Reading the result depends on understanding the scope split. Scope 1 captures emissions from sources the infrastructure operators control directly, essentially fuel combusted on site in backup generation. Bakers, rollup operators and data availability nodes run on general-purpose servers in commercial data centers and offices rather than on dedicated industrial equipment, so direct combustion attributable to the network is negligible and the scope 1 figure is reported as effectively nil. Scope 2 captures the indirect emissions embodied in the electricity those servers purchase and accounts for practically the whole footprint. It is derived on a location basis from average grid intensity, not on a market basis, because renewable energy certificates or power purchase agreements held by individual operators leave no trace in network data and cannot be verified from outside.

The boundary excludes emissions embodied in manufacturing and eventually disposing of the hardware, and it excludes the electricity used by wallets, indexers, explorers and applications built on the chain, which are counted against those services rather than the network. Carbon intensities are annual averages, so seasonal and daily variation in a grid's mix is not represented, and every uncertainty in locating nodes carries straight through into the emissions result.

Greenhouse gas intensity is stated as a marginal figure, the additional emissions attributable to one further transaction at the present node population and throughput; because the consumption behind it is close to fixed, that figure falls as activity rises and should not be read as an efficiency measurement. Carbon intensity values are taken from Carbon intensity of electricity generation, compiled by Our World in Data from Ember and the Energy Institute's Statistical Review of World Energy and published under the CC BY 4.0 license.