What is staking? A plain-English guide that covers the parts most explainers skip
Staking is not a savings account. It is a bonded job with a 32 ETH deposit, a slashing budget, and a withdrawal queue that hit 850,000 ETH in July 2024. Here is the version of the explainer most people skip.
On 15 September 2022, at slot 4,700,013, Ethereum stopped paying miners and started paying validators. The Merge cut the network’s energy use by roughly 99.95% overnight, according to the Ethereum Foundation’s own measurements, and it turned 32 ETH from a unit of account into a job application. Three years and one Shanghai upgrade later, the staked-ETH pool sits north of 34 million ETH, the active validator set has crossed one million, and the conversation around “staking” has split into two things that no longer mean the same thing: the protocol-level job of attesting to blocks, and the consumer product that wraps it.
That gap — between what the protocol is paying you for and what your provider is selling you — is where every interesting question about staking lives. It is also where most explainers stop. This piece picks up there: the bonded-job framing, the slashing budget, the withdrawal queue that hit 850,000 ETH after Shanghai, the difference between solo and pooled and liquid, and why “yield” is the wrong word for the number on the screen.
Staking is a bonded job, not a savings account
The cleanest way to picture an Ethereum validator is as a notary on retainer. You post 32 ETH as a bond. Every 12-second slot you might be asked to attest to a block, and every 32 slots — one epoch, six minutes and 24 seconds — your committee is rotated. If you do the job correctly, the protocol pays you a base reward funded by issuance plus a share of priority fees and any MEV your block included. If you sign two conflicting attestations, or if you propose two blocks at the same slot, the protocol burns part of your bond and ejects you. That penalty is called slashing, and it is the only reason any of this works.
The retainer framing matters because it explains the economics. A validator is not lending capital. They are renting it to the consensus protocol as collateral against misbehaviour. The yield is closer to an insurance premium than to interest, and it scales with risk: more validators means more competition for the same issuance, so the per-validator number drops. The issuance curve sits at roughly 2.6% gross for a solo validator at one million active validators, and drifts toward 2.3% as the set grows. After the realised burn from EIP-1559, net issuance has spent most of 2024 and 2025 in the slightly negative range — meaning ETH is mildly deflationary even as stakers are paid.
The slashing budget nobody quotes you
Slashing is the part the marketing pages skip. A standalone slashing event costs the validator a minimum of 1 ETH (about 1/32nd of the bond) plus an “inactivity leak” if they go offline at the same time. The bad case is correlated slashing, where many validators run the same client, the same signing setup, or the same cloud region and get caught by the same bug at the same time. The protocol’s correlation penalty scales quadratically with how much of the validator set is slashed in the same 36-day window. At 1% of the set, the marginal penalty is roughly 3% of stake; at 33%, it is the entire bond. Ben Edgington’s annotated spec remains the readable reference for the math.
In practice, slashings are rare. Beaconcha.in’s running tally shows fewer than 500 slashings against more than a million validators since genesis, and almost all of them were operator errors: a redundant signing key brought online by mistake, a backup node accidentally promoted to primary, a botched migration. The lesson is not that slashing is theoretical — it is that the slashing budget is the price of the rare bad day, and any honest staking provider should publish theirs. Most do not.
Solo, pooled, liquid — three different products
The retail experience of staking comes in three flavours that share a name and very little else. The differences are not nuance; they are the entire product.
| Model | Minimum | Counterparty risk | Withdrawal time | Typical net APR (2026) |
|---|---|---|---|---|
| Solo (32 ETH, own hardware) | 32 ETH | None beyond client bug | Subject to exit queue | ~3.1% |
| Solo via DVT (Obol, SSV) | 32 ETH spread across operators | Operator set | Exit queue | ~2.9% |
| Pooled (Rocket Pool minipool) | 8 ETH + RPL bond | Pool smart contract + node operator | Withdraw on-demand via rETH | ~2.7% |
| Liquid staking token (Lido stETH) | Any | Lido DAO + 40+ node operators | ~1-5 days via withdrawals queue | ~2.6% |
| Centralised exchange (Coinbase, Binance, Kraken) | Any | The exchange | Exchange policy | ~2.2-2.4% after fees |
The product that has eaten the market is liquid staking. Lido alone holds roughly 9.4 million ETH — about 28% of all staked ETH — and stETH is the most common collateral asset across Aave, Maker, and Spark. Rocket Pool is a distant second with around 1.1 million ETH and a fundamentally different design: each node operator posts a 10% RPL bond against their minipool, which makes the network resemble a permissionless co-op rather than a curated whitelist. Coinbase’s cbETH sits around 1.5 million and is the largest centralised LST. See our market dashboard for the running tally.
The withdrawal queue and why Shanghai changed everything
Before April 2023, staked ETH was one-way. The Shanghai/Capella upgrade switched on withdrawals, and the queue immediately became the most-watched metric in the staking economy. Two things sit in that queue: partial withdrawals, which sweep accumulated rewards every few days and are throttled per epoch; and full exits, which return the 32 ETH bond and are throttled by the consensus-layer churn limit. The churn limit currently allows roughly 8 full exits per epoch, which works out to about 1,800 validators per day, or 57,600 ETH.
That ceiling matters during stress events. After the post-Shanghai unlock in 2023, the exit queue peaked at roughly 850,000 ETH and took most of June to clear. The reverse — the entry queue — has hit similar numbers twice, including the run-up to the Pectra activation. Anyone holding an LST should understand that “instant” liquidity from a token like stETH or rETH is really a secondary-market price; the underlying ETH still has to walk through the same queue, and the LST can trade at a discount to its redemption value when the queue is long. The famous June 2022 stETH dislocation hit a 7.5% discount; it returned to par in weeks, but only because withdrawals were eventually switched on.
Where the “yield” actually comes from
The number on a staking dashboard is usually three numbers smashed together: protocol issuance, priority fees, and MEV. Issuance is the steady part — the per-validator share of the issuance curve. Priority fees are the gwei users pay above the EIP-1559 base fee, and they are highly variable; during quiet weeks priority fees can add 30-50 basis points to APR, during NFT mints or memecoin frenzies they can briefly double the headline number. MEV is the third component, captured almost entirely through MEV-Boost, and it is the one most retail products mark up the hardest.
- Issuance: ~2.3-2.6% gross, predictable, set by the validator-count curve.
- Priority fees: ~0.2-0.6% on average, spikes during congestion. Track current levels on our gas dashboard.
- MEV via MEV-Boost: ~0.3-0.5% on average, fat-tailed, occasionally dominated by a single sandwich block.
- Operator commission: Lido 10%, Rocket Pool 14% on the node-operator half, Coinbase 25%.
Two things follow from this breakdown. First, “yield” is the wrong frame: you are not earning interest on ETH, you are earning a share of network revenue denominated in ETH, plus an issuance subsidy. Second, the gap between the highest and lowest advertised APRs is almost entirely a fee story, not a performance story — the underlying validators are doing the same work. Anyone comparing offers should look at the post-fee number and the slashing-coverage policy, in that order.
Tax, custody, and the boring parts that decide returns
Three operational details quietly decide whether staking actually beats holding spot. The first is tax. In the United States, the IRS Revenue Ruling 2023-14 treats staking rewards as ordinary income at fair-market value on the date of receipt, with cost-basis carried into any subsequent sale. In the UK, HMRC has consistently treated rewards as miscellaneous income or trading income depending on activity level. In Germany, individual stakers who held for more than one year used to enjoy tax-free disposal, though recent guidance has tightened the holding period for staked assets to retain the exemption.
The second is custody. A validator’s withdrawal credentials are the single most important key the operator holds. Pre-Capella, many providers used contract-controlled withdrawal addresses (0x01 credentials); post-Pectra, the new 0x02 credentials enable partial withdrawals from balances above 32 ETH, which changes how operators consolidate stake. If you are using a custodial product, read the credential type. The third is client diversity. The execution-layer client split is finally healthier than it was in 2023 — Geth is below 50% for the first time, with Nethermind, Erigon, and Besu sharing the rest — but the consensus-layer side is still dominated by two clients. Use the clientdiversity.org tracker before choosing an operator.
What to actually do with this
The honest answer for most readers is: pick the product that matches your tolerance for counterparty risk, accept that the post-fee APR will be between 2% and 3%, and stop chasing the last 30 basis points. For balances above 32 ETH where you want to keep custody, distributed validator technology — Obol, SSV — is now production-grade and removes the single-node failure mode that has historically slashed solo stakers. For balances below 32 ETH where you want full decentralisation, Rocket Pool minipools and the new Lido CSM (Community Staking Module) are the two paths that actually deepen the validator set rather than concentrating it.
If you are mainly here for the number, the difference between the best and worst offering is usually 40-60 basis points after fees, which is real money on a six-figure stake and rounding error on a four-figure one. Either way, the part that decides whether you keep the principal is not the APR — it is whether your operator has client diversity, a slashing-coverage policy, and a public incident history. Those are the questions to ask. Run the math on a few scenarios in our staking calculator before you commit, and watch upcoming consensus-layer upgrades on the events calendar — the next protocol change to issuance will move every number in this article.
Staking is not a savings account. It is a bonded job, and the people doing it well are the ones who treat it like one.