Running a validator is not difficult once. It is difficult continuously, across client upgrades, key rotations, and the 3am failure that only becomes expensive if the failover is configured wrong. Most slashing events in the history of proof-of-stake have not been attacks. They have been a duplicate key running on two machines.
What the job actually is
On Ethereum, a validator stakes 32 ETH, runs a consensus client and an execution client, and attests to the chain's state every epoch. Rewards come from correct attestations and from proposing blocks; penalties come from being offline (small, proportional to how many others are also offline) and from provable misbehaviour (large, and designed to be).
The daily work is unglamorous: apply client updates on a schedule because every one of them is also a potential consensus-split bug, monitor disk and peer count, watch for missed attestations, and keep the signing key somewhere that is not the same place as the failover node's configuration.
The duplicate-key slash, and how to prevent it
Ethereum's slashing conditions are narrow and severe: double voting (signing two different blocks at the same height) and surround voting (voting around another validator's vote). Both are produced by a key signing two conflicting messages, and the most common cause is infrastructure rather than malice.
Do not run two active nodes with the same key. This is the mistake that produces most slashing events. It usually happens accidentally — a provider's failover starts signing while the primary is still alive, or a migration leaves the old node running.
A correct failover is not automatic. The safe patterns are primary-active with a backup that only monitors, delayed automatic failover with manual confirmation, or a threshold signature scheme requiring multiple parties to authorise signing. Every one of these trades recovery speed for the guarantee that two signers cannot exist at once.
Watch the exit path too. On Ethereum, an exiting validator sits in a withdrawal queue measured in days, during which it is still running and still exposed. An operator who assumes an exit is instant will keep nodes alive longer than intended.
What the provider landscape costs you
Three tiers, and each gives up something different.
Custodial providers — the exchange staking products — run the validator inside an account you control on their platform. Frictionless, and the stake is in someone else's custody. You also lose most of your governance ability, since the provider votes the stake.
Non-custodial services run the validator for you while you keep the withdrawal keys, taking a commission that typically lands in the 5–15% range of rewards. You retain ownership and lose some control over the operator's infrastructure; your keys are not at risk, but the operator's uptime is your uptime.
Liquid staking issues a receipt token representing the stake — Lido, Rocket Pool — which adds DeFi composability and removes the operator decision entirely: you choose a token, not a validator. What you give up is the choice itself, plus smart-contract risk, the possibility of the receipt token trading below its peg, and a concentration question that has become the live controversy in the space: how much of the network ends up under a handful of operators.
How to evaluate an operator
The question worth asking is not "what is your uptime" — everyone claims the same number — but "what happens when your primary node fails."
Specifically: does the failover share the signing key with the primary, and is there a mechanism that guarantees only one can sign at a time? How long does failover take, and is it automatic or confirmed? What happened the last time an infrastructure provider had an outage? And what is the operator's slashing history, verified rather than reported?
For a delegator the same questions apply with an extra one: on liquid staking, does the token's price reflect a slashing event immediately, and what is the protocol's policy when it does? The answer determines whether you are insulated from an operator's mistake or exposed to it.
The honest summary
Staking is a business with an operational tail. The rewards are modest, the risks are concentrated in a small number of configuration decisions, and the failure mode is a single event that costs a large fraction of the stake. If you are not going to run the infrastructure with the discipline the numbers require, a provider is the correct answer — provided you ask what happens when their primary node dies.
References
Next, read The Restaking Ecosystem 101. More in Infrastructure & Scaling.