"How do I stake Ethereum" sounds like one question, but it's really four. There are four distinct routes into staking — running your own validator, delegating to an operator, joining a pool, or letting an exchange handle it — and they differ in almost everything that matters: how much ETH you need, who holds the keys, what can go wrong, and how you get out. Most of the confusion around staking comes from advice written for one route being applied to another.
The underlying mechanism is the same in every case, though, and it's worth understanding before choosing a door. Whichever route you take, your ETH ends up backing a validator — and the validator's behavior, not your choice of interface, is what determines the risk.
What Staking Actually Does
Ethereum is secured by proof of stake. Validators — software processes with 32 ETH or more posted as collateral — take turns proposing blocks and attesting to blocks proposed by others. Do the job correctly and the protocol pays rewards from new ETH issuance, plus priority fees and MEV from the transactions in proposed blocks. Do it dishonestly — sign two conflicting blocks, for instance — and the protocol destroys a portion of the collateral and ejects the validator. That penalty is slashing, and it's the enforcement mechanism that makes the whole system work.
Two things follow from this that affect every staking route. First, rewards float. The protocol pays more per validator when fewer validators exist, and less as more ETH stakes — the rate is a function of total participation, not a promise. Anyone quoting a fixed yield is describing a snapshot, not a feature. Second, penalties are graded. Going offline costs roughly what you would have earned — annoying, not catastrophic. Slashing is reserved for provable double-signing, which in practice comes from misconfiguration (running the same validator keys in two places) rather than malice.
The Four Routes
Solo staking means running the validator yourself: 32 ETH deposited to the protocol's deposit contract, an execution client and consensus client running on hardware you control, and your own keys. It's the only route with no intermediary — rewards arrive at the protocol rate, minus nothing. The costs are the 32 ETH threshold, the operational duty (the machine needs to stay online and patched), and key management with no one to call. One detail worth knowing: the validator key that signs duties is separate from the withdrawal credentials that control the funds. A compromised validator key can get you penalized; it can't steal your stake.
Staking-as-a-service keeps the 32 ETH requirement but delegates the operations: you hold the withdrawal credentials, an operator runs the validator for a fee. You're trusting their competence — their misconfiguration is your slashing — but not their honesty with your principal, provided the withdrawal credentials genuinely stay yours. That proviso is the thing to verify, not assume.
Pooled and liquid staking removes the threshold. Protocols like Lido and Rocket Pool aggregate deposits, spread them across operators, and issue a token (stETH, rETH) representing your staked position. Any amount works, and the token stays liquid — you can sell it rather than queue for a protocol exit. The trade is a new layer of risk: smart contract bugs, operator set quality, and the fact that the liquid token trades at a market price that can deviate from the underlying ETH during stress. It usually tracks closely. "Usually" is doing real work in that sentence — stETH traded at a visible discount during the 2022 deleveraging, and that episode is the honest reference point for what stress looks like.
Exchange staking is the custodial route: the exchange stakes on your behalf, takes a cut of rewards, and handles everything. Operationally it's the simplest, and it carries the full custodial trade — the exchange holds the keys, you hold a claim on the exchange. The staking risk and the counterparty risk stack.
How You Get Out
Exits deserve more attention than they usually get. Since the Shapella upgrade, staked ETH is withdrawable, but not instantly. Reward balances above 32 ETH sweep out to withdrawal addresses automatically. A full exit means joining the validator exit queue, which the protocol deliberately throttles — under normal conditions it clears in hours to days, but it's sized to slow a mass exodus, because validators leaving en masse is exactly the scenario the protocol needs to resist. Liquid staking tokens sidestep the queue by letting you sell the position instead, which is genuinely useful, with the caveat already noted: the market price during a rush for the exit is the one moment it's least likely to be at par.
What's Changing
The Pectra upgrade in 2025 changed the validator math. EIP-7251 raised the maximum effective balance from 32 to 2,048 ETH, so large operators can consolidate thousands of validators into fewer, larger ones, and solo stakers' rewards now compound toward the higher cap instead of idling. EIP-7002 made exits triggerable from withdrawal credentials — meaning someone using a service operator no longer depends on that operator's cooperation to exit. That's a quiet but real reduction in delegated-staking trust. Meanwhile the share of staked ETH sitting in liquid staking protocols and restaking systems keeps the concentration question alive: the mechanism is permissionless, but the stake distribution is lumpier than the design assumed.
What Would Confirm or Break This Picture
Confirmation: exit queues continuing to clear in expected timeframes through volatile periods, liquid tokens holding close to par outside acute stress, and post-Pectra consolidation proceeding without consensus incidents.
Invalidation: a slashing event caused by a protocol bug rather than operator misconfiguration would break the "penalties are predictable" claim. A liquid staking token failing to recover its peg after stress passes, or an exit queue extending to months under load, would break the liquidity assumptions that make the pooled routes attractive.
Timing Perspective
Now: All four routes are functional and withdrawable; the route choice is a trust-model choice, not a timing one. Next: Watch how validator consolidation under the 2,048 ETH cap plays out, and whether delegated services adopt credential-triggered exits as standard. Later: Restaking layers on top of staked ETH are still proving out — they change the risk of the position, not the mechanism described here.
Boundary
This explains the mechanism behind each staking route and where the trust sits in each. It is not a recommendation to stake ETH, a yield forecast, or an endorsement of any protocol, operator, or exchange named above. Tax treatment of staking rewards varies by jurisdiction and isn't covered here. Whether staking makes sense for any particular holder depends on circumstances outside this post's scope — the static explanation lives here; the tracked version lives elsewhere.
