Concentration risk in supply, stake and validators
A network can be decentralized in software and concentrated in practice, and ownership, operation and governance each concentrate separately.
Concentration risk is the possibility that a small number of parties can determine an outcome that the system is described as distributing. It appears in at least three independent places: who owns the units, who operates the machines that produce blocks, and who decides what the rules are. A network can be widely distributed on one of these dimensions and tightly held on another, so a single decentralization claim is rarely meaningful without asking which one is meant.
Supply concentration and what address counts do not show
Ownership concentration begins with distribution at launch. An insider allocation to founders, early investors and a foundation is normal, and what matters is its size relative to total supply and the schedule releasing it. Future released supply is described by an unlock schedule and summarized on this site as dilution overhang, which expresses supply not yet in circulation as a share of the total. That figure is a measure of how much can arrive, not a claim about when or whether it will be sold. Concentration is also difficult to observe directly, because addresses are not people. A custodian holding assets for thousands of clients appears as one address, while a single holder can split a balance across hundreds. Counts such as addresses with a balance and top-holder tables are therefore upper and lower bounds rather than measurements, and are best read as such.
It is worth separating two things that the word concentration covers. Ownership concentration is a market fact: a small number of holders can move a price when they transact, and their intentions are unobservable, but they cannot take anyone else's units or change any rule. Operating and governance concentration is a control fact: a small group can halt finality, reorder transactions, or alter the rules that govern assets already deposited. The first affects what a position is worth and the second affects whether the system behaves as described. Both are commonly reported under the same heading, and conflating them leads to conclusions that do not follow, since a widely held asset can run on tightly held infrastructure and the reverse is equally possible.
Stake concentration on proof-of-stake networks
Where security depends on staked value, the distribution of that stake is the distribution of influence over block production and finality. Several measures apply. The staking ratio shows how much of the supply is bonded, which affects both security and the share of supply available to trade. The number of validators counts the entities producing blocks, which overstates independence when one operator runs many. Delegation means holders can assign their stake to operators without transferring ownership, so operating concentration can be far higher than ownership concentration. Liquid staking providers concentrate this further by aggregating stake from many holders behind one protocol and a set of operators it selects, and trading venues that stake customer assets do the same thing through a different structure. The consequence is correlated failure: an operator error, a hosting outage or a client bug affecting one large operator can lead to slashing or downtime across a large share of stake simultaneously, which is a different risk from any individual validator misbehaving.
Software and infrastructure concentration
A network running one dominant implementation carries a risk that is invisible in every ownership metric. If a supermajority of validators run the same client and that client has a consensus defect, the defect becomes the network's behavior, and the honest minority is the side that appears to be wrong. Node client diversity is the measure of this, and it is one of the few risks the community that runs the network can address directly by choosing differently. Infrastructure concentration works the same way one layer down: validators hosted in one cloud provider, one region, or one jurisdiction share an outage and share a legal exposure. A published node count tells how many machines participate and nothing about where they are or what they run.
Block production, ordering and censorship
Even with well-distributed stake, the construction of blocks can concentrate. On networks where specialized parties assemble blocks for validators, a small number of builders and relays can end up producing most blocks, which gives them influence over transaction ordering and therefore over MEV extraction, and creates a point at which particular transactions could be excluded. Censorship risk in practice is less about a network refusing transactions than about a concentrated set of intermediaries applying the same policy at the same time, with inclusion delayed rather than denied. On proof-of-work networks the analogous concentration is in mining pools: individual miners are numerous, but hash power is directed by a small number of pool operators, and a coalition controlling a majority could reorganize recent history, which is the 51 percent attack that smaller networks have actually experienced.
The Nakamoto coefficient and what it leaves out
The Nakamoto coefficient counts the minimum number of entities that would need to cooperate to control a chosen subsystem, most often block production or stake. It is the most useful single summary available and carries three caveats. It depends entirely on which subsystem is measured, so a network can score well on stake and poorly on clients or on bridge control. It counts declared entities, and two entities under common ownership or sharing an infrastructure provider count as two. And it is a threshold measure, so it says nothing about the distribution below the cutoff. Reading it alongside client diversity, hosting distribution and the identity of the largest operators produces a more accurate picture than any of them alone.
Data quality, covered next, determines how far any of the figures above can be trusted in the first place. The staking pages show stake distribution per network, supply covers ownership and unlocks, and methodology explains how each measure is constructed.