Stake concentration measures how much of a network’s bonded capital sits with a small number of operators. It is usually discussed as a governance or ideology topic. It is also a yield topic, because concentration decides who produces blocks, who gets paid, and what happens to your rewards when one operator stops performing.
High stake concentration does not reduce your nominal rate on day one. It changes the distribution of uptime, the durability of the reward schedule and the tail risk attached to your position. Those effects are slower and larger than the headline share suggests.
- Concentration is measured by entity, not validator. One operator running twenty validators is one point of failure.
- It reaches your yield through variance. The average holds; the bad quarters get worse.
- Thresholds matter more than averages. A third and two thirds of stake are the numbers that change behaviour.
- It is checkable. Every input is public on a block explorer and takes minutes to assemble.
What stake concentration measures
The measure is the share of total bonded stake held by the largest operators, counted as entities rather than as validators. The distinction is the whole point: a network with 150 validators and 8 operators is far more concentrated than its validator count implies.
Three metrics that answer different questions
Top-one share tells you whether a single operator can act alone. Top-five and top-ten shares tell you whether a small group could coordinate. The Nakamoto coefficient tells you how many entities would have to act together to cross a threshold, usually a third or two thirds of stake.
Each number is useful and none is sufficient. A chain with a modest top-one share and a heavy top-ten share is not decentralised; it is just decentralised at the first level of inspection.
Why validator counts mislead
Validator counts are easy to inflate and easy to display. Delegation programmes, exchange staking products and institutional custodians often run many validators under one policy, one key management regime and one set of infrastructure. When that regime has a bad night, all of them miss blocks together.
| Metric | What it captures | Where it fails |
|---|---|---|
| Validator count | Size of the active set | Ignores common ownership |
| Top-one share | Whether one entity can act alone | Misses coordination among the next few |
| Top-ten share | Control by a small group | Hides how the group is connected |
| Nakamoto coefficient | Entities needed to cross a threshold | Depends entirely on the threshold chosen |
| Stake by custodian | Who really controls the keys | Rarely disclosed, hard to verify |
How concentration reaches your yield
Concentration does not change the reward formula. It changes the inputs to that formula, and it changes how often they misbehave at the same time.
Correlated downtime
Independent operators fail independently, so the network loses a little stake at random. Concentrated operators fail together, so the network loses a lot at once. Your expected return over a year barely moves; the variance around it widens considerably.
An operator holding 38% of stake with 1.5% downtime removes about 0.57% of network rewards for that period. If the same 38% were split across forty independent operators with the same average uptime, the expected loss would match but the chance of a simultaneous outage would be far lower.
Pricing power over commission
When a few operators control most of the stake, commission stops being competitive. Delegators have limited credible alternatives, switching costs a redelegation or an unbonding period, and the largest operators can raise rates in step without losing much stake.
That is the direct yield link. Concentration does not tax you through the protocol; it taxes you through the reduced bargaining power you have over the people who run it.
Security spend competing with your rewards
A network whose stake is concentrated has to pay more, in issuance, to achieve the same security, because the marginal cost of attacking it is lower. That trade-off is settled at the protocol level and shows up as either higher inflation or a thinner security margin, both of which land on delegators eventually.
Decentralisation is usually argued as a principle. It is easier to defend as an insurance premium: the price of not having your yield depend on one operator’s upgrade schedule.
Stake concentration and the finality threshold
Thresholds turn concentration from a preference into a mechanism. On most proof-of-stake networks, finality needs more than two thirds of stake participating, and a group holding more than a third can prevent finality by withholding.
Take a network with 120 active validators where the top four operators hold 9.5% each, 38% combined. The one-third line is crossed by four entities: 33.4 divided by 9.5 is 3.5, rounded up to 4. The Nakamoto coefficient at that threshold is 4.
For the two-thirds threshold the same arithmetic gives 8 entities. That is the number that can finalise blocks without anyone else, and it is worth knowing before you treat the network’s validator count as evidence of resilience.
Now price that into a delegation decision. If your chosen operator sits inside the top four, a coordinated outage does not just cut your rewards, it stops the chain from finalising, which freezes exits at exactly the moment you would want one.
Checking stake concentration before you delegate
- Pull the validator list from a block explorer and export stake per validator.
- Group validators by operating entity, not by name or by moniker. Shared infrastructure, shared funding and shared governance are the tells.
- Calculate top-one, top-five and top-ten shares of total bonded stake.
- Compute the Nakamoto coefficient at one third, and again at two thirds.
- Check how much of the top group is reachable through the staking interface you actually use, since that is the set you can influence.
- Note your chosen operator’s position in that ranking, and what share of the network its outage would remove.
- Repeat the exercise quarterly, because delegation flows move faster than governance.
0 comments
No comments yet — ask a question and we will answer it here.