On 3 March 2020, accounts belonging to three exchanges, Huobi, Binance and Poloniex, cast votes on the Steem blockchain. According to Decrypt’s report the next day, the voting power behind them came from their customers’ STEEM. The votes handed control of the chain to block producers backed by Justin Sun, who had bought the company Steemit a few weeks earlier. Steem ran on delegated proof of stake, and that day showed what the design rests on: whoever holds the coins holds the vote, even when the coins belong to someone else.
Delegated proof of stake (DPoS) is a way of running a blockchain in which token holders do not produce blocks themselves. They vote, with their tokens as weight, for a short list of operators who do. Tron uses it today with 27 Super Representatives; EOS, renamed Vaulta, uses 21 block producers. Many other networks borrowed the word delegation while building something different. On Tezos, Cosmos Hub, Polkadot, Cardano and BNB Chain, holders also hand their weight to an operator, but who produces blocks, and who loses money when an operator misbehaves, follow other rules.
This lesson follows our lesson on validators, rewards and slashing, which covered what an operator earns and how penalties work. Two reminders before going further. In proof of stake as explained in depth, validators lock coins and their influence is proportional to that stake. In our introduction to consensus mechanisms, DPoS appears as a branch of proof of stake in which power comes from the backing of token holders. This lesson opens that branch, down to what a holder actually gives away when ticking delegate in a wallet.
One word, three designs: what delegated means
The word covers at least three arrangements. Mixing them up leads to wrong answers to simple questions such as whether a holder can lose tokens.
The first is DPoS in its original sense: holders elect a fixed, small number of block producers. BitShares, Steem and its fork Hive, EOS (now Vaulta) and Tron belong here. The number of producers is a protocol constant, 101 on BitShares, 21 on EOS, 27 on Tron. Producers take turns on a schedule, and every holder can change their vote at any time.
The second is stake delegation inside an ordinary proof-of-stake system. The protocol ranks validators by the stake bonded to them, their own plus what holders delegate, and the top of the list becomes the active set. Cosmos Hub caps that set at 180 validators and BNB Chain at 45. There is no ballot in the formal sense, but validators still compete for delegators on commission and reputation.
The third is a group of hybrids that add a rule of their own. Tezos lets holders delegate without locking anything. Polkadot runs an election algorithm meant to spread stake evenly across validators. Cardano pays pools less once they grow past a target size.
One thing is common to all of them: on these chains the holder keeps the tokens in their own wallet and hands over only weight. What changes is how that weight becomes block production, and who carries the risk when something goes wrong.
Where DPoS came from: Daniel Larimer and BitShares
BitShares: 101 delegates and a block every 10 seconds
DPoS is credited to Daniel Larimer, the developer behind BitShares and later Steem and EOS. Accounts of when he first described it differ, from late 2013 to 2014, and sources do not agree on the date of its first deployment either. The clearest dated text is his own: a blog post of 7 January 2015 titled The Most Decentralized Proof of Stake System.
In that post Larimer writes that BitShares “elects 101 delegates with approval voting of the shareholders”. Approval voting lets each holder back several candidates at once instead of picking one. The 101 delegates were shuffled at random every 101 blocks, then took turns producing a block every 10 seconds.
His case was economic. Running a block producer costs money, and he argued that anyone holding less than 1 percent of the stake could not produce blocks profitably anyway. Small holders would do better lending their weight to a professional than running a node, and an election lets them do so while keeping the producers replaceable at any time.
That reasoning set the template for everything that followed: few producers, a fast fixed schedule, and accountability by vote rather than by financial penalty. It also planted the main criticism, because a vote counted in tokens can be bought, pooled or borrowed.
Steem and EOS: the same model at larger scale
Larimer reused the model in Steem, a blogging chain whose block producers were called witnesses, and then in EOS, built by the company Block.one. The EOS mainnet started in June 2018 with 21 block producers. Block.one’s token sale had raised what the SEC called “the equivalent of several billion dollars over approximately one year”. On 30 September 2019, the SEC announced that Block.one would pay a 24 million dollar civil penalty to settle charges that the sale was an unregistered securities offering, without admitting or denying the findings.
EOS was renamed Vaulta in 2025 and kept the design. Vaulta’s own guides say every account can vote for up to 30 candidates, with votes weighted by the tokens the voter has staked, and the top 21 by vote weight produce blocks. A block comes every 0.5 seconds, in rounds of 126 in which each producer signs six blocks in a row. A block becomes irreversible once 15 of the 21 producers, two thirds plus one, have confirmed it. The guides describe no slashing: a producer that misses its turn is skipped, and repeated misses can lead to removal or demotion.
Tron’s 27 Super Representatives: DPoS in daily use
How the Tron vote works
Tron is the clearest live example of the original model. Tron’s developer documentation sets out the steps: to vote, an account first stakes TRX to obtain TRON Power, then spreads that power across candidates. At each maintenance period, every six hours, the network re-ranks the candidates by votes.
The top 27 become Super Representatives and produce blocks in rotation. Coindaro measured an average of 3.0 seconds per block over the 19,273 blocks up to block 87,819,273 on 8 November 2026, read from the public Trongrid API. Ranks 28 to 127 are Super Representative Partners: they produce no blocks but share in voter rewards and can submit proposals. Below that, an account is simply a candidate.
The money side, from the same documentation and from the chain parameters read through Trongrid on 8 November 2026: 8 TRX for each block produced, plus 128 TRX per block shared among the top 127 in proportion to their votes. By default a Super Representative keeps 20 percent of that voter reward and passes 80 percent to the accounts that voted for it; each one can change the ratio. Unstaked TRX becomes available again after 14 days. The documentation describes no slashing for producers or voters, and all these values can be changed by on-chain proposals.
Who holds the votes on 8 November 2026
Coindaro read the full list of 446 witnesses through Trongrid’s listwitnesses call at about 09:44 UTC on 8 November 2026. It counted 43.87 billion votes in total. The names below are each operator’s self-declared web address, not a verified identity, and the shares are shares of all votes in the list.
| Rank | Registered name (web address field) | Votes | Share of all votes | Cumulative share |
|---|---|---|---|---|
| 1 | Poloniex | 4,859,660,036 | 11.1% | 11.1% |
| 2 | t.network | 3,708,849,094 | 8.5% | 19.5% |
| 3 | tronalliance.org | 3,046,129,283 | 6.9% | 26.5% |
| 4 | JDinvestment | 2,743,544,673 | 6.3% | 32.7% |
| 5 | CryptoLabs | 2,535,800,661 | 5.8% | 38.5% |
| 6 | blockanalysis | 2,198,569,174 | 5.0% | 43.5% |
| 7 | Huobi wallet | 1,911,293,767 | 4.4% | 47.9% |
| 8 | cryptoinnovationfund | 1,717,439,480 | 3.9% | 51.8% |
| 20 | OKX earn | 858,858,762 | 2.0% | |
| 24 | Binance staking | 784,062,273 | 1.8% | |
| 26 | okcoin.jp | 724,763,603 | 1.7% | |
| 27 | Kraken staking | 611,821,214 | 1.4% | 96.9% |
| 28 (first partner) | trxultra.org | 248,448,890 | 0.6% | 97.5% |
Two readings follow. By votes, five names hold more than a third of everything cast and eight hold more than half. And six of the 27 producers carry the web address of an exchange or of an exchange’s staking product: Poloniex, Huobi, OKX, Binance, OKCoin Japan and Kraken. Behind their votes sit, at least in part, customers who staked through the platform and never picked a candidate themselves.
Block production is spread differently. Each of the 27 producers gets the same share of slots, whatever its vote count. In the terms of the Byzantine generals problem, a group controlling more than a third of the producers can stop agreement, so 10 of the 27 if each is truly independent. Votes decide who sits in those 27 seats, and that is why the distribution above matters.
Vote buying and the cartel critique: EOS in 2018
EOS, autumn 2018: allegations of mutual voting
The objection to DPoS came early. If producers are paid and voters choose them, producers have a reason to pay voters, and to vote for one another. EOS gave the first public test. In late September 2018 a document circulating on Chinese social media alleged “collusion, mutual voting, and pay-offs” among EOS block producers, SiliconANGLE reported on 1 October 2018. The document singled out the exchange Huobi, alleging that it and other large producers voted for one another to keep their seats, and that Huobi received EOS in exchange for votes. Huobi denied the allegations.
Block.one’s chief executive, Brendan Blumer, wrote that the company was aware of “unverified claims regarding irregular block producer voting”. The EOS Alliance, a community group, told members that vote buying should be handled “without brand-damaging drama, witch hunts, or mob rule”. The claims were never proven and remain allegations.
Why the economics push producers together
The incentive problem does not depend on that one case. A producer seat pays a steady income. A voter who receives part of it, as the default 80 percent voter share does openly on Tron, is in effect paid for voting. On Tron that is public and part of the rules, which makes the line between sharing rewards and buying votes hard to draw. Producers can also vote their own tokens for one another, turning a set of competitors into a group with a shared interest in staying elected.
Exchanges add a second problem. A platform holding tokens for millions of customers can vote with all of them, and those customers rarely see how. Approval voting makes the effect larger: when one stake can back many candidates at once, a single large holder can weigh on most of the seats at the same time.
Steem and Hive, March 2020: when exchanges’ coins vote
Day by day
In February 2020 Justin Sun bought Steemit, the company behind the Steem blogging platform, and with it the 65 million STEEM it held, worth about 12 million dollars at the time according to Decrypt. Part of the community feared that this stake would be used to vote. In late February, nine days before Decrypt’s report of 4 March, the community’s witnesses approved a soft fork that blocked those tokens until Sun explained his plans.
On 3 March 2020, accounts run by Huobi, Binance and Poloniex used their STEEM voting power to give Sun’s side control of the chain. Binance and Huobi withdrew their votes on 4 March, and Binance’s then chief executive Changpeng Zhao wrote on Twitter that Binance had “no interest in chain governance. We stay neutral.” Poloniex issued no statement, Decrypt reported.
Sun gave his own account. He wrote that he wanted to protect private property and the interests of all from “malicious hackers”, according to Decrypt. Critics called it a takeover; his side called it a defence. Neither version has been tested in court.
On 18 March, community members announced a hard fork, The Block reported. CoinDesk had described the plan on 17 March: everyone holding STEEM would receive the same amount on the new chain, except the wallets controlled by Steemit, whose development-fund tokens would not be carried over, and staked tokens on the new chain could not take part in governance votes for 30 days, a brake on any repeat of the exchange vote. The new chain, Hive, went live on 20 March 2020 as a copy of Steem’s state. Steem kept running under Sun’s side; Hive became a separate network with its own token.
What the Steem episode proved
Nothing in the protocol broke. Every vote was valid and every block was correctly signed. The design assumed that the tokens voting belonged to people with a stake in the chain’s future, and custodial exchanges broke that assumption. The community’s only defence after the fact was to copy the chain and leave.
The lesson reaches beyond DPoS. In any vote weighted by tokens, whether for block producers or for a protocol upgrade, the question to ask is who holds the tokens, and then how many voted.
Delegated proof of stake variants, chain by chain
Tezos: liquid proof of stake and its bakers
Tezos calls its block producers bakers. According to the protocol constants read through the TzKT API on 8 November 2026, a baker needs at least 6,000 tez of its own, blocks come every 6 seconds, and a cycle of 14,400 blocks lasts about a day. Tezos names its design liquid proof of stake, because the ordinary holder’s weight moves without the tokens being locked.
A holder has two options, as the Tezos documentation on staking sets out. Delegating assigns the tez to a baker’s weight while the tokens stay fully under the holder’s control and can be spent at any time; the docs say it incurs no risk, and it earns only a fraction of the rewards. Staking goes further: the tez stay in the holder’s account but cannot be transferred, and they face the same penalties as the baker’s own stake, in proportion to their share. Unstaking takes up to four days, three cycles plus the rest of the current one.
Bakers also carry governance. The Tezos governance docs state that only delegates vote on protocol proposals and that their voting power equals their staking balance, which includes what holders delegated to them. On 8 November 2026, TzKT counted 201 active bakers with a combined staking balance of 667.3 million tez, 140,620 delegators and 3,304 stakers.
Cosmos Hub: a capped set of 180 validators
The Cosmos Hub, the chain of the ATOM token, ranks validators by bonded stake and keeps the top 180. The staking parameters read from the chain on 8 November 2026 give a maximum of 180 validators, an unbonding time of 1,814,400 seconds, or 21 days, and a minimum commission of 5 percent. Bonded stake stood at 340.63 million ATOM.
Agreement among those 180 runs on CometBFT, the Tendermint lineage that a later lesson of this path covers. What matters for a holder is the risk: on the Cosmos Hub delegators share their validator’s penalties. The percentages and the unbonding mechanics are in the previous lesson on slashing, and they apply in full to someone who delegates.
Polkadot: nominators and the Phragmén election
Polkadot calls its design nominated proof of stake. Holders become nominators and back a list of validators; the chain then runs an election, built on a method devised by the Swedish mathematician Lars Edvard Phragmén, that aims to maximise the total stake behind the active set while keeping the stake of each validator as even as possible. The result is fewer giant validators and fewer starved ones than a simple ranking would give.
The Polkadot wiki’s chain-state page, read on 8 November 2026, lists 600 active validators, a minimum bond of 250 DOT to nominate and 28,161 nominators. Eras last about 24 hours. The same page still gives an unbonding period of 28 days, but for nominators that figure is out of date.
Until mid-2026, Polkadot was one of the chains where backers carried the operator’s risk: when a validator was slashed, the active nominators behind it lost a share too, and leaving took 28 days. Referendum 1910 changed that. Polkassembly lists it as executed, and the Polkadot forum announced on 6 July 2026 that it had been enacted: “Nominators are no longer slashable”, and their unbonding period fell from 28 days to about two days, or two eras. Validators were left out of the change. They keep a minimum self-stake, remain slashable on their own stake and keep their longer unbonding and slash-deferral periods. Unbonding requests started before the upgrade stayed on the old 28-day schedule. Polkadot’s developer docs, last updated on 29 June 2026, still describe nominator slashing as the rule and the change as expected, so pages written before July still describe the old rules.
Cardano: pools, k and pledge
On Cardano, holders delegate to stake pools by registering a delegation certificate, which costs an ordinary transaction fee, plus a refundable deposit the first time a stake address is registered. Cardano’s documentation says the ada can be spent normally at any time, whatever the delegation, and it describes no slashing for delegators.
The size of pools is steered by a parameter called k. Cardano’s protocol parameters, read through the Koios API for epoch 666 on 8 November 2026, give k = 500 and a pledge influence of 0.3. According to Cardano’s developer curriculum, a smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool reaches saturation at the total stake divided by k; past that point, rewards per ada fall, which pushes delegators toward smaller pools. Pledge, the operator’s own ada in the pool, raises rewards, so an operator with nothing at stake earns less. Active stake in epoch 666 was 21.33 billion ada. Some explainers still cite k = 200, a worked example in the docs, not the current value.
BNB Chain: Proof of Staked Authority
BNB Chain blends DPoS with proof of authority. Its documentation states: “The active validator set consists of 45 validators: 21 Cabinet and 24 Candidates.” Candidates are ranked by bonded BNB, including delegations, and holders can delegate and re-delegate. After the Lorentz, Maxwell and Fermi upgrades, the mainnet produces a block about every 0.45 seconds, and blocks reach fast finality in about one second once two thirds of validators have voted.
The variants side by side
| Network | Block producers | How they are chosen | Delegator slashed? | Exit time | Source, date |
|---|---|---|---|---|---|
| BitShares (2015) | 101 delegates | Approval voting by holders | No penalty described | n/a | Larimer, 7 Jan 2015 |
| EOS / Vaulta | 21 | Stake-weighted votes, up to 30 candidates each | No slashing described | Not stated in guides | Vaulta guides, Nov 2026 |
| Tron | 27 (127 rewarded) | TRON Power votes, re-ranked every 6 hours | No slashing described | 14 days | Tron docs, Trongrid, 8 Nov 2026 |
| Tezos | 201 active bakers | Baking power (own stake, staked and delegated tez) | Delegators no; stakers yes | Delegation: none; staking: up to 4 days | Tezos docs, TzKT, 8 Nov 2026 |
| Cosmos Hub | 180 | Top by bonded stake | Yes | 21 days | Cosmos Hub chain, 8 Nov 2026 |
| Polkadot | 600 | NPoS election (Phragmén method) | No, since 6 July 2026 (validators still yes) | About 2 days for nominators | Polkadot wiki, forum, referendum 1910, Nov 2026 |
| Cardano | Stake pools (k = 500) | Delegation; saturation at total stake / k | No slashing described | None for delegation | Koios, Cardano docs, 8 Nov 2026 |
| BNB Chain | 45 (21 Cabinet, 24 Candidates) | Top by bonded BNB | Not covered here | Not covered here | BNB Chain docs, Nov 2026 |
How concentrated is delegated stake? Three live readings
The Nakamoto coefficient, and the threshold you pick
The usual yardstick is the Nakamoto coefficient: the smallest number of entities that together control enough of the network to cause trouble. The answer depends on the threshold. More than a third is the point where, in a BFT system, a group can stop the chain from finalising blocks. More than half is the point of simple majority control. A published coefficient means little unless it says which threshold it uses.
The count also depends on what one entity is. One company can run several validators under different names, and an exchange’s validator carries mostly its customers’ stake. Published cross-chain rankings differ from one provider to the next for those reasons, so Coindaro computed its own figures from public chain data for three networks, on the same morning.
Tron, Cosmos Hub and Tezos on 8 November 2026
| Network | What was counted | Set size | Largest single share | Entities for more than 1/3 | Entities for more than 1/2 |
|---|---|---|---|---|---|
| Tron | Votes, Trongrid listwitnesses | 27 producers (446 listed) | 11.1% (Poloniex) | 5 | 8 |
| Cosmos Hub | Bonded ATOM, Cosmos Hub REST | 180 validators | 18.7% | 4 | 8 |
| Tezos | Baker staking balance, TzKT | 201 bakers | 15.8% (Coinbase Baker) | 3 | 7 (5 if the Tezos Foundation counts as one) |
On Tezos the two largest bakers carry the labels Coinbase Baker (105.18 million tez, 15.8 percent) and Kraken Baker (69.21 million, 10.4 percent), followed by Everstake and Kiln. Three accounts labelled Tezos Foundation Baker hold 73.8 million tez between them, 11.1 percent. Counted as one entity, the foundation brings the number needed for a majority down from seven to five. On the Cosmos Hub, a single validator held 18.7 percent of bonded stake; four reached a third and eight reached half.
These readings come with caveats. The labels are what the operators or the explorers call them, not audited ownership. Most of the stake behind a large validator belongs to delegators, who can leave, at once on Tezos for plain delegation, after 21 days on the Cosmos Hub, after 14 days on Tron. And Tron’s vote count is not the same thing as its block production, which is shared equally among the 27. Still, three or four names hold more than a third of the stake on Tezos and the Cosmos Hub, enough to stall finality if they acted together, and on Tron five names hold a third of the votes that choose the producers. Several of those names are exchanges.
What delegating means for a holder
What stays with you and what leaves
On the chains above, delegating does not hand your tokens to the operator. The operator receives weight, not keys, and cannot spend what you delegated. What varies is how free your tokens remain. Plain delegation on Tezos and delegation on Cardano leave them spendable at any moment. On Tron the TRX must be staked to vote and takes 14 days to come back after unstaking. On the Cosmos Hub, bonded ATOM needs 21 days to unbond. On Polkadot a nominator’s DOT has needed about two days since July 2026, down from 28.
Rewards flow the other way. The operator earns from the protocol, keeps its commission (a minimum of 5 percent on the Cosmos Hub, a default 20 percent on Tron) and passes the rest to its backers. A commission can change, so the rate on the day you delegate is not a promise.
What can still go wrong
The first risk is the penalty. On the Cosmos Hub, and on Tezos for holders who stake rather than delegate, a delegator’s tokens are cut when the chosen operator is slashed. Polkadot nominators were in the same position until July 2026 and no longer are. On Tron, Vaulta and Cardano no slashing is described, but an operator that goes offline or loses its seat earns nothing for its backers, and the same now goes for a Polkadot validator that misbehaves.
The second is the exchange route. Staking through a trading platform is not delegation in the sense above: the platform holds the keys, chooses the validator and casts the vote, as the Steem episode showed. The third is outright fraud. Websites promising fixed staking returns on deposits are a common cover for theft, and our guide to fake crypto investment platforms lists the warning signs. The step-by-step method for staking and delegating safely comes in a later lesson of this path.
Rules around delegation in the US, the UK and the EU
For US readers, the reference text is a staff statement of the SEC’s Division of Corporation Finance published on 29 May 2025 on what it calls Protocol Staking Activities. Its view is that these activities “do not involve the offer and sale of securities” within the meaning of the Securities Act of 1933. It covers solo staking, self-custodial staking with a third-party node operator, where a footnote mentions owners who “delegate” their validation rights to a node operator, and custodial arrangements. It excludes liquid staking, restaking and liquid restaking, and it says that, like all staff statements, it “has no legal force or effect”. It does not discuss producer elections of the Tron or Vaulta type as such.
Tax is a separate question. Under IRS Revenue Ruling 2023-14, a cash-method taxpayer who stakes the native coin of a proof-of-stake chain includes the rewards in gross income at their fair market value when gaining “dominion and control” over them, and the same applies when staking through an exchange; our guide to US crypto taxes covers the reporting. In the UK, HMRC’s Cryptoassets Manual treats staking rewards outside a trade as miscellaneous income at their sterling value when received, with capital gains tax on a later disposal. The manual does not discuss delegation separately; our UK crypto tax guide gives the detail.
For EU readers, ESMA’s Q&A 2067, answered by the European Commission on 20 June 2024, says MiCA has no staking-specific provision and “does not therefore prohibit staking”. Staking services offered to clients are treated as ancillary to custody, so the provider needs a MiCA authorisation for custody, and losses from staking are to be deemed attributable to the provider. The answer concerns firms that hold clients’ assets; it does not deal with a holder delegating from their own wallet.
What DPoS trades away for speed
The bargain is visible in the numbers. A small, known set of producers can agree quickly: Vaulta signs a block every 0.5 seconds and makes it irreversible once 15 of 21 producers confirm, BNB Chain reaches finality in about a second, Tron produces a block every three seconds. Fewer machines also mean lower costs per transaction.
The price is paid in who decides. Votes weighted by tokens follow wealth and custody, exchanges vote with customers’ coins, and producers have reasons to support one another. The 2018 EOS allegations and the 2020 Steem episode are the two cases most often cited, and the concentration readings above show how few names sit at the top even on chains that never used the DPoS label.
The variants are attempts at a fix. Polkadot’s election evens out stake, Cardano’s k penalises oversized pools, Tezos lets holders leave a baker at any moment, and the Cosmos Hub widens the set to 180. None removes the basic fact that weight follows tokens. How a capped set of validators actually reaches agreement, round by round, is the subject of the next lesson of this path, on BFT consensus with Tendermint and HotStuff.
