Polkadot (DOT) sustainability report

Polkadot is present on the following networks: Polkadot.

Polkadot runs a Nominated Proof of Stake system on its relay chain, and it deliberately separates producing blocks from declaring them irreversible. Holders of the native asset may bond it and either stand as a validator candidate or act as a nominator, naming a short list of candidates they are willing to back. At the start of each era an on-chain election algorithm derived from Phragmén's methods — a sequential variant and a maximin-support refinement of it — picks the active validator set from the candidates and also decides how each nominator's bond is divided among the validators it named. The aim is not merely to seat the largest stakes but to even out the backing behind elected validators, so no single validator concentrates a disproportionate share and the cost of capturing the set stays high. The size of the active set is a governance parameter rather than a fixed constant.

Block production is handled by BABE. Time is cut into epochs of six-second slots, and for every slot each validator privately evaluates a verifiable random function against the epoch's shared randomness; an output below a stake-weighted threshold entitles that validator to author the slot. Two validators can win the same slot, briefly forking the chain; when nobody wins, a deterministic fallback fills it so authoring never halts. Finality is a distinct mechanism, GRANDPA, running alongside the chain BABE builds. Validators vote on chains rather than individual blocks, and once more than two thirds of the active set have voted for a chain containing a given block, that block and all of its ancestors are finalized in one step. The interaction is loose by design: authoring proceeds at a steady rate whatever finality is doing, and finality can confirm a long run of blocks in a single round rather than gating production.

Parachains do not elect validators of their own. Their collators assemble candidate blocks with a proof of validity, a small group of relay chain validators re-executes that proof and backs the candidate, and the candidate data is erasure coded across the whole validator set so any large enough subset can reconstruct it. Further validators then self-select by lottery to re-check the work; a contradictory result opens a dispute that the entire set must settle, and validators on the losing side are slashed. Only candidates that survive this approval stage are finalized.

Polkadot is present on the following networks: Polkadot.

Validators are paid once per era out of newly issued units of the network's native asset. Payment follows era points, awarded for authoring relay chain blocks and for validation work performed on parachain candidates, so a validator with a larger bond does not automatically out-earn a diligent smaller one. Each validator first takes the commission it advertises; the remainder is shared between its own bonded stake and the stake nominators placed behind it, in the proportions the era's election assigned. Nomination pools let smaller holders combine bonds and back validators collectively. Rewards are claimed rather than pushed out, and unclaimed payouts expire after a set number of eras.

Penalties distinguish serious faults from ordinary unreliability. Equivocation — authoring or voting for two conflicting things in the same slot or round — and backing a parachain candidate that proves invalid are slashable, and the slashed fraction rises with the number of validators committing the same offense in the same window, so an isolated fault costs far less than a coordinated one. Nominators behind a slashed validator forfeit a matching share of their bond. Downtime alone is not slashed; a persistently unresponsive validator is dropped from the active set and stops earning until it is restored. Bonded funds released by a validator or nominator sit through an unbonding delay before they can be moved and stay exposed to slashing for offenses committed beforehand. These staking parameters are governance-controlled and have been revised more than once.

Users pay a weight-based fee. Every call carries a weight expressing the execution time and storage access it is expected to need, and the charge is a fixed base amount, plus an amount for the encoded length of the transaction, plus an amount derived from that weight; an optional tip buys queue priority. A multiplier tracks how full recent blocks have been and moves the weight-to-fee conversion up or down between blocks within a bounded range, so sustained congestion raises costs gradually instead of spiking. Roughly four fifths of the inclusion fee is directed to the on-chain treasury and the remainder to the block author. Storing data on chain requires refundable deposits returned when the data is released. Blockspace itself is no longer won at auction: a chain buys coretime, either as a bulk region covering a fixed span or on demand for a single block, and the native asset spent on it is burned.

Polkadot is present on the following networks: Polkadot.

The figure is assembled from the machines that actually keep the network running, not from any price or revenue signal. Polkadot's consensus rewards no computational effort, so there is no mining hardware to model; what draws electricity is the population of relay chain validators together with the collator and full node infrastructure of the chains that share its security. Sizing that population comes first. Validator counts are read directly from on-chain state, while the wider node population is estimated from network crawlers and publicly visible peer data, producing a reachable-node count that is then adjusted upward for nodes that do not advertise themselves.

A representative hardware profile is inferred for each class of participant from the reference specification the client software asks operators to meet — processor class, memory, and storage type — on the reasoning that operators provision close to what the client demands and rarely far beyond it. Power draw for those device classes is taken from controlled bench measurement rather than from manufacturer nameplate ratings, and the measurement covers idle draw as well as draw under load, since a validator spends much of each slot waiting rather than executing. Total consumption is the aggregate over the estimated node set at those measured rates across the reporting period.

Because the relay chain provides security to the chains connected to it, the line between one chain's consumption and another's is an attribution question rather than a measurement one. A connected chain's figure is its own collator and node infrastructure plus a share of the relay chain validator set, apportioned by the relay chain resources that chain consumes. Where an asset exists on more than one network, the network totals are combined according to observed on-chain transfer activity for that asset.

None of this is metering. The node count, the hardware mix and the utilization rate are inferred from public observation and from stated software requirements, and each is an estimate carrying its own error. Where the evidence is thin, the assumption selected is the one producing the higher number, so the result is more likely to overstate the footprint than to understate it. Figures are revised as crawler coverage and hardware measurement improve, and successive reporting periods are therefore not always directly comparable.

Polkadot is present on the following networks: Polkadot.

The renewable share is derived geographically rather than from any contractual claim about the electricity the operators buy. The starting point is where the network's machines physically sit. Node locations are inferred from publicly observable network data — the addresses peers advertise, routing information, and the hosting providers those addresses resolve to — and aggregated to the country level, since finer resolution would be spurious and would expose operator detail without improving the estimate. Hosting concentration is treated with care: a large share of nodes sitting in a handful of data center regions is a real feature of the distribution, not a sampling artifact to be smoothed away.

Where the geographic spread cannot be observed directly for part of the node set, a structurally similar network stands in for the missing portion. Similarity here means a comparable participation and reward design and a comparable consensus family, on the reasoning that networks which attract operators on similar terms tend to attract them in similar places. The substitution is applied to the unobserved remainder only, not to the whole distribution.

Each country weight is then matched to that country's generation mix, giving a renewable share for the electricity the network's infrastructure consumes. The public dataset used for the generation mix is Share of electricity generated by renewables, compiled by Ember and the Energy Institute's Statistical Review of World Energy and processed by Our World in Data. Because the mix is an annual national average, it reflects the grid an operator draws from rather than any specific supply arrangement that operator may hold, and short-term or seasonal variation within a country is not captured.

Energy intensity is reported separately and means the marginal energy cost of one additional transaction: the change in consumption attributable to adding one transaction to the load the network already carries, rather than total consumption divided by transaction count. On a network whose validator set draws power continuously regardless of how busy it is, the two are very different quantities, and the marginal figure moves with throughput even when the underlying infrastructure has not changed at all.

Polkadot is present on the following networks: Polkadot.

Emissions follow from the energy estimate and the same geographic picture, applied to grid carbon intensity instead of generation mix. Node locations are inferred from publicly observable network data and aggregated to the country level; where part of the node set cannot be placed, the distribution of a structurally similar network — one with a comparable participation and reward design — substitutes for the unobserved remainder. Each country weight then carries that country's average carbon intensity of electricity generation, and the weighted result is applied to the energy figure to produce the emissions total.

The reporting distinguishes two scopes. Scope 1 covers emissions from sources the network's operators directly control — combustion on site, refrigerant loss, on-premises generation — and for a network of this kind it is effectively nil, because the infrastructure is servers in third-party facilities rather than anything that burns fuel. Scope 2 covers the indirect emissions embodied in the electricity purchased to run that infrastructure, and it accounts for essentially the whole figure. Emissions further upstream, such as those from manufacturing and shipping the hardware or from constructing the data centers themselves, sit outside both scopes and are not included here.

The carbon intensity data is drawn from Carbon intensity of electricity generation, compiled by Ember and the Energy Institute's Statistical Review of World Energy with substantial processing by Our World in Data, and made available under a CC BY 4.0 license. These are location-based annual averages: they describe the grid serving a country, not any supply contract or renewable certificate a particular operator may hold, so a facility running on contracted clean power is not distinguished from its neighbours.

GHG intensity is the marginal emission attributable to one additional transaction, computed on the same basis as the energy intensity figure. Its uncertainty compounds that of every input beneath it — the node population, the hardware profile, the inferred locations and the national intensity averages — and it should be read as an order-of-magnitude indicator rather than a precise per-transaction measurement.