On a normal April 2026 day, roughly 917,000 validators each hold 32 ETH of bond and take turns signing messages on Ethereum's Beacon Chain. Together they secure about 35 to 39 million ETH and earn around 3.0% APR for it, according to ethereum.org's live staking dashboard. That headline hides a stack: an attestation protocol running on 6.4-minute epochs, a block-production market routed through MEV-Boost, a withdrawal pipeline that only opened in 2023, a 2025 upgrade that changed the economics of running a node, and a shadow market of liquid staking tokens and restaking protocols that most ETH holders actually interact with instead of running their own validator.
This article walks the stack from the Beacon Chain up to EigenLayer. If you only want to know which staking product to click, skip to Section 5 — but the trade-offs there only make sense once you know what the validator is doing underneath.
What a Validator Actually Does
To activate a validator, you deposit 32 ETH into the deposit contract and register two cryptographic keys: a BLS signing key (hot, used every epoch) and a withdrawal credential (cold, controls where rewards and exits land). Since Shapella enabled withdrawals in April 2023, the withdrawal credential should be an execution-layer address — the 0x01 prefix — rather than the older BLS-keyed 0x00 format that requires a one-time BLSToExecutionChange migration.
Once active, a validator has three duties:
- Attest every epoch. Each epoch is 32 slots of 12 seconds — 6.4 minutes — and each validator casts three votes: source, target, and head. These votes are weighted 14, 26, and 14 out of a total reward budget of 64.
- Propose blocks when selected. Roughly once every few months for an average validator, the protocol randomly picks you to propose. Proposer weight is 8 out of 64, but the real upside here is that block proposers also collect priority fees and MEV tips.
- Serve on sync committees. A rotating subset of 512 validators signs lightweight headers that light clients rely on. Sync-committee weight is 2 out of 64.
In practice, about 90% of proposers outsource block building to MEV-Boost relays, signing whatever payload the external builder returns in exchange for a cut of the block's extracted value. That is where most of the yield above the pure-issuance APR comes from: priority fees flow through MEV-Boost to validators, directly into the proposer's fee-recipient address rather than being minted as new ETH.
Rewards, Penalties, and the Slashing Knife-Edge
Issuance rewards are inversely proportional to the square root of total active balance — the more ETH staked network-wide, the less each individual validator earns. That is why APR has drifted down as deposits have grown. Reward math splits into the five weighted components above (source, target, head, sync, proposer). Miss an attestation and you forfeit those weights for the epoch; miss enough and you start to accumulate penalties.
Two penalty regimes are worth understanding because they are qualitatively different:
Inactivity leak. If the network fails to finalize for four or more consecutive epochs — meaning a supermajority of validators can't agree — inactive validators' balances quadratically decay until finality is restored. This is the protocol's recovery mechanism during a mass outage, not a punishment for brief downtime. Ordinary offline blips cost you a few missed rewards, not a meaningful chunk of stake.
Slashing. Slashing fires only on equivocation: signing two blocks in the same slot, casting a double vote, or surround-voting a prior attestation. The initial penalty burns 1/32 of your effective balance — 1 ETH on a 32 ETH validator. You are then forced into an exit, and during a 36-day window the protocol applies a correlation penalty proportional to the total slashed balance across the network. If you are the only validator slashed that month, the total penalty is modest. If 10,000 validators running the same misconfigured client are slashed together, the correlation penalty can approach your entire stake. Slashing is not really a punishment for bad acts — it is an insurance mechanism that makes correlated failures expensive for operators running homogeneous setups.
Deposits, Withdrawals, and the Exit Queue
The Shapella upgrade in April 2023 turned ETH staking from a one-way deposit into a round-trip. Two kinds of withdrawals came with it:
- Partial withdrawals happen automatically via the withdrawal sweep: the protocol scans up to 16,384 validators per sweep and pays out any balance above the cap (originally 32 ETH, now higher — see the next section) into the
0x01address. Up to 16 withdrawals can land per block. - Full withdrawals require the validator to explicitly exit, wait out the exit queue, then wait an additional withdrawability delay before funds are released.
The exit queue is rate-limited to prevent coordinated dumping. Since Pectra, the churn limit is denominated in ETH (~256 ETH per epoch, roughly 57,600 ETH per day) rather than validator count, which cleanly handles the new variable-balance world. During the September 14, 2025 congestion spike, end-to-end exit-to-withdrawability stretched to 46+ days — a useful data point when someone tells you LST discounts are "arbitrage."
Pectra Changed the Validator Picture
The Pectra upgrade activated at epoch 364032 on May 7, 2025, and it meaningfully reshaped validator economics. Three EIPs matter for stakers:
- EIP-7251 (MaxEB): raises
MAX_EFFECTIVE_BALANCEfrom 32 ETH to 2048 ETH while leaving the 32 ETH minimum to activate untouched. Large operators can consolidate up to 64 redundant validators into a single one via the consolidation contract at0x0000BBdDc7CE488642fb579F8B00f3a590007251, rate-limited to one to two consolidations per block. Solo stakers get something too: rewards above 32 ETH now compound on the validator instead of being swept out and needing to be redeposited. - EIP-7002: lets the execution-layer withdrawal address trigger a validator exit. That means the cold key you already use to receive funds can also pull the ripcord, without needing the hot signing key. For serious operators, this is a meaningful safety upgrade.
- EIP-6110: delivers deposits natively from execution to consensus instead of the old eth1data poll, cutting activation latency.
Net effect: the operational picture got strictly better for careful stakers.
Solo vs SaaS vs LST: A Trade-off Map
Most ETH holders don't run validators. The real decision is which staking product to use, and every option is fundamentally a choice about which counterparty and slashing risks you accept. This is the section worth reading twice.
Solo / home staking is the protocol's intended default. You hold the 32 ETH, you run the node, you own both keys, you eat the uptime cost. You also keep 100% of rewards and MEV. The risk is narrow and personal: your setup, your slashing liability, your electricity bill. If you are technical and hold enough ETH, this is the closest match between what the protocol rewards and what you experience.
Staking-as-a-Service (SaaS) providers like Kiln, Figment, or Allnodes run the node for you while you retain the withdrawal keys. You are trusting the operator for uptime and for not getting you slashed, but you are not trusting them with your funds — the withdrawal credential still points at your address. A meaningful middle ground for large holders who don't want to run infrastructure. Tax treatment remains personal.
Liquid staking tokens (LSTs) pool your ETH with others and hand you a transferable token. LSTs are a crypto-backed asset class comparable to stablecoins: the token is a claim on the underlying staked position, and its market price can diverge from its redemption value.
The three benchmarks:
- Lido stETH is the dominant LST. stETH rebases daily — your balance grows. wstETH is the non-rebasing wrapped variant used in most DeFi integrations because rebasing tokens don't play well with lending pools. Lido charges 10% of rewards, split evenly between node operators and the DAO. Lido's share of all staked ETH sits around 24% in early 2026, down from a ~32% peak in late 2023. The one-third threshold matters because a single entity above that line can theoretically influence finality, which is why Lido's decline below 25% is treated as meaningful news rather than market share chatter.
- Rocket Pool rETH uses an exchange-rate model (the token price appreciates rather than the balance changing). Operators bond 8 ETH (LEB8 minipools) plus 2.4 to 12 ETH of RPL collateral — 10% to 150% of the borrowed amount — and the remaining 24 ETH is matched from the rETH deposit pool. The operator set is permissionless, which is a structurally different governance profile than Lido's curated operator list.
- Coinbase cbETH is a centralized LST: one issuer, one jurisdiction, one set of validators. The exchange-rate model and the operational simplicity are attractive to some holders; the concentration and regulatory risks are the obvious trade-off.
Every LST collapses the 46-day exit queue into "sell on the market," but at the cost of discount risk. When 3AC and Celsius were unwinding in May and June 2022, stETH traded as much as ~7% below ETH on secondary markets — not because staking was broken, but because forced sellers flooded a thin redemption path. If you borrow against an LST, plan for that scenario.
Restaking: EigenLayer and the Layered Risk Stack
Restaking reuses an already-staked position — either natively via an EigenPod contract set as your 0x01 credential, or indirectly by depositing an LST — as additional collateral for "actively validated services" (AVSs): oracles, bridges, data availability layers. EigenLayer hit roughly $18-19.5 billion in restaked TVL and 1,900+ operators by early 2026, with EigenDA as the largest consumer.
The honest framing is important. Restaking is not free yield. You are selling additional slashable insurance on a second protocol, and you are doing so in the same collateral you already have at risk to Ethereum. Dual slashing means a single operator bug can trigger both Ethereum's slashing conditions and an AVS's — potentially correlated, potentially costly.
Liquid restaking tokens (LRTs) like ether.fi eETH, Renzo ezETH, and Kelp rsETH wrap restaked positions into another transferable token. You now have three layers: validator, restaked position, LRT wrapper. Restaking layers new counterparty and slashing risk in a composable stack, and the failure modes are not additive — they can correlate. If an AVS slashes or an LRT de-pegs, LSTs used as lending collateral can trigger liquidations before you have time to react. The April 2026 Kelp DAO bridge exploit — which drained roughly $292 million in rsETH and forced Aave, Compound, and other venues to freeze rsETH markets — is a reminder that the LRT layer itself is an attack surface.
Restaking can make sense for operators who would run the infrastructure anyway and understand the AVS risk profile. For passive holders, it is a yield bump that comes with a genuinely different risk grade than plain staking — treat it that way.
Key Takeaways
- A validator is a 32 ETH bond with three duties — attest, propose, sync — whose rewards decompose into five weighted components totaling 64.
- Slashing is insurance against correlated failure, not a penalty for offline time; the 36-day correlation window is what makes homogeneous operator setups dangerous.
- Shapella enabled withdrawals in 2023; Pectra in 2025 raised MaxEB to 2048 ETH, let execution-layer addresses trigger exits, and redenominated the exit queue in ETH.
- Choosing between solo, SaaS, and LSTs is a choice about which counterparty, centralization, and discount risks you accept. Lido's ~24% share and the 46-day exit queue are both reasons that choice matters.
- Restaking and LRTs layer new slashing conditions on top of staking — closer to insurance underwriting than to additional yield.
Part 5 turns to Ethereum's scaling layer — how rollups translate blob data back into user-facing cheap transactions.
