Unstaking Ethereum: The Address You Only Get to Set Once

Most explanations of unstaking Ethereum answer one question: how long does it take. That framing assumes the money is behind a single door with a queue in front of it.

It is not. Rewards leaving a validator and the validator itself closing down are different mechanisms, and a third question sits underneath both. That question is which withdrawal credential type the validator carries. It is decided once, cannot be undone, and determines whether anything arrives on its own at all.

What Actually Leaves a Validator, and When

A validator account lives on the consensus layer, spendable ETH on the execution layer. A staking withdrawal is the movement between those halves, and three things can cause it.

The first is a partial withdrawal: reward balance leaving while the validator stays active. The second is a full withdrawal, which happens after the validator has exited and sends everything remaining. The third is not a movement but a precondition, and it is the one that traps people.

That precondition is a registered withdrawal address. Without one, neither withdrawal can happen, because the protocol has nowhere to send anything.

Withdrawals of any kind only became possible with the Shanghai and Capella upgrade on 12 April 2023, and Pectra in May 2025 added a second credential type. Guidance written between those dates describes a network with one kind of validator, and much of it is still online.

Type 1 and Type 2 Credentials Are Not a Detail

Every validator carries withdrawal credentials of one of two types, marked by a prefix. Type 1 begins 0x01 and belongs to a legacy validator; Type 2 begins 0x02 and belongs to a compounding one. A third prefix, 0x00, is an absence rather than a type: no withdrawal address is attached and withdrawals are not enabled at all.

The difference between the two live types is a ceiling. A legacy validator has an effective balance capped at 32 ETH, and anything above that does nothing for it. A compounding validator can carry up to 2048 ETH, and rewards fold back into that balance in whole-ETH steps, held back by a small buffer that stops the figure flickering.

From that one difference follows the part that reverses a claim repeated almost everywhere.
Rewards on a legacy validator leave every few days with no transaction and no fee at any step, which is where the line about automatic, cost-free rewards comes from.

On a compounding validator the same sentence is false. Nothing is swept until the balance clears 2048 ETH, so taking reward balance out below that means triggering a partial withdrawal by hand from the execution layer, and that transaction costs gas.

The pattern is not that one is better. They are opposite defaults: one assumes the excess should leave and moves it without asking, the other assumes it should stay at work until told otherwise.

The Two Decisions You Only Get to Make Once

Two choices here cannot be reversed, and neither announces itself as permanent when it is made.
The first is the withdrawal address. A validator accepts one, once, and after it reaches the consensus layer it cannot be undone or swapped for another.

The reason is structural. A withdrawal address points at a smart contract or a key-controlled account, and neither can send a message back to the consensus layer asking for the credentials to be repointed.

There is a way to keep flexibility, but it has to be taken at the start. Naming a smart contract wallet that can rotate its own keys lets the eventual recipient change later, because the registered address never moves.

An operator who registered a plain key-controlled account and later loses confidence in it has one slow route left: exit, let the balance land at the registered address, then activate a new validator with different credentials.

The second irreversible decision is the conversion from Type 1 to Type 2. Raising a validator from the 32 ETH ceiling to the 2048 ETH ceiling is one-way, with no route back.

That conversion is also what makes consolidation possible, since several validators can be merged into one, and the sweeping behaviour changes permanently along with it.

Why Nothing Has Arrived: The Three Questions a Proposer Asks

When a validator is chosen to propose a block, part of its job is to look through validator accounts and decide which are owed a withdrawal. Each account is put through the same short test.

The first question is whether a withdrawal address has been provided. If none has, the account is skipped, nothing is initiated, no error is raised and nothing is held over for later.

The second is whether the validator has exited and reached the point where its account counts as withdrawable. If it has, the whole remaining balance goes to the withdrawal address.

The third is whether the balance sits above the ceiling for its credential type, 32 ETH for a legacy validator and 2048 ETH for a compounding one. If it does, and the account is active with its effective balance already at maximum, only the excess is transferred.

Turned around, that sequence is a diagnostic. A staker seeing nothing arrive is failing exactly one of three tests, and they can be checked in order. In practice the first is the usual culprit, for anyone who deposited before withdrawals existed and never registered an address.

ETH in that state is not at risk, assuming the seed phrase is still safe, but it does not move either. Only two operator actions influence any of this: providing credentials switches withdrawals on, and exiting triggers the full one.

The Sweep Is a Rotation, Not a Queue You Join

The word queue does damage here, because it suggests requests lining up in the order they were made. Nothing of the sort happens for partial withdrawals.

Proposers work through validator indices in numerical order without skipping any. Each picks up where the last stopped, builds a list of up to 16 eligible withdrawals from that point onward, and the next continues from there. At the highest index, the pass restarts at zero.

An account is never waiting for a turn it requested. It waits for the pass to reach its index, and that pass runs whether or not anything is owed.

The arithmetic follows from one limit. At most 16 withdrawals fit in a block, which is 115,200 in a day if no slots are missed, and a full cycle takes as long as walking past every validator account ever registered.

Over 1.2 million had been registered as of April 2026, though accounts with nothing eligible are passed over quickly, which pulls the real interval below the worst case.

Exiting Fully, and Why the Wait Has No Fixed Number

Leaving staking altogether starts with a voluntary exit, and there are two ways to submit one.
The choice decides what has to be in your hands, and what it costs.

The first route signs the exit message with the validator client and broadcasts it to a consensus node, and no gas is involved because nothing touches the execution layer. The second triggers the exit from the execution layer using the withdrawal credentials, without needing the signing key at all, and that is a transaction, so it costs gas. That cost is not one number either, and what a gas fee is made of decides how much of it comes back.

Once the exit is submitted the validator waits, and that wait depends on how many others are leaving at the same time. It is a fact about network conditions, not a property of your account, which is why a printed figure is worth little. A number in an article was true the day it was measured; the value that matters is on a beacon chain explorer at the moment you exit.

When the exit completes the account stops having duties, stops earning and is marked withdrawable. From there the next sweep to reach its index sends the entire balance onward.

One consequence surprises people afterwards. An exited validator cannot be revived by depositing more ETH into it; anything sent there is swept out on a later pass. Staking again means activating a new validator.

What This Changes Before You Stake Rather Than After

Almost all of this is easier to act on before the deposit than after it. The withdrawal address is the clearest case: whether the recipient should be a plain account or a contract wallet that can rotate keys belongs to setup, not to the day of the withdrawal.

The credential type is the second, since planning to compound and consolidate later is choosing a different withdrawal experience, not only a different ceiling.

There is also a liquidity point with no mechanics in it. ETH sitting in a validator is supply that is locked rather than circulating, and it cannot be treated as available on demand.

That matters most where crypto is used as collateral. Staked ETH is not capital that can be moved quickly when leveraged positions come under pressure, since the exit wait is set by other people leaving rather than by need.

Traders arriving from derivatives sometimes read staking rewards as a yield comparable to the periodic payment between holders of opposing positions. The two are unlike: one is paid by the protocol for work performed, the other is a transfer between traders.

And opening an account with an exchange to stake through it puts none of the above in your hands. The provider holds the validator relationship and its timetable governs what arrives.

Who Should Not Run Their Own Validator

Running a validator at home gives the most control and the fewest intermediaries, and carries obligations that are easy to underestimate. It starts with 32 ETH to activate the validator and a machine running an execution layer client and a consensus layer client continuously, both needing upgrades when the network changes.

Rewards are proportional to the time the validator is online and attesting properly. Downtime is penalised on a sliding scale: the cost rises when many validators are dark together and falls when yours is one of the few. Recommended bandwidth is at least 10 Mb/s each way.

Downtime penalties are not slashing, and separating the two is worth doing. Slashing is heavier and reserved for offences the protocol reads as malicious, with the risk minimised by running a minority client and loading the keys on one machine only.

Three profiles should probably not run their own validator: anyone unable to commit to maintenance and network upgrades, anyone without 32 ETH they are content to lock, and anyone who could not keep validator keys and seed phrases secure unaided.

The alternatives trade control against that burden. A staking service still requires the 32 ETH deposit but removes the hardware, at the cost of sharing signing keys. Pooled arrangements accept small amounts with no keys and no hardware, giving up the direct protocol relationship instead.

Whichever route is taken, the risks attaching to the asset do not change. They are set out in the risks that come with holding crypto, and locking it in a validator adds a constraint rather than removing one.

Frequently Asked Questions