Restaking Is Overbuilt, and Somebody Is About to Lose a Lot of Money

The restaking industry is now $50 billion of pooled ETH backstopping services that, in most cases, could not be identified by a randomly sampled restaker if you held a gun to their head. The yields are in the 3-6% range above ETH's native staking rate. The implicit assumption, priced into those yields, is that the additional slashing risk is essentially free money. It isn't.

What an AVS actually is

An Actively Validated Service is a piece of software whose correctness is backed by the slashable capital of EigenLayer-restaked validators. In theory, the slashing condition is specific to the AVS's defined misbehaviors. In practice, most AVSes are:

• Running on open-source software with limited formal verification
• Operated by small teams with modest budgets for security engineering
• Secured against slashing losses by insurance products that have never been tested at scale
• Paying yields that do not appear to reflect the actual correlation risk across services

The correlation argument

The critical point — which the Ethereum Foundation's risk panel has been making for two years, to almost no practical effect — is that a validator restaking across multiple AVSes is not taking on independent risks. The risks are correlated in ways that the yield math does not reflect.

A single bug in a widely-used middleware library that underlies many AVSes would slash validators across all affected services simultaneously. A consensus failure in a shared sequencer would slash across its dependents. The individual AVS slashing probabilities might each be low. The joint distribution — the probability of being slashed on at least one service given that you are slashed on any — is much higher than the sum of the marginals implies.

"Everyone is priced as though they are taking uncorrelated risks. The risks are not uncorrelated. I have been saying this. I will continue to say this." — Justin Drake, Ethereum Foundation

The yield is not compensation

If you believe, as I do, that the correlation risk is meaningfully higher than the yield premium compensates for, then the restaking market is a negative-expected-value bet that the entire industry has collectively agreed to pretend is positive-expected-value. The pretend is sustained by the absence of a large slashing event. Every additional month without one calcifies the misperception.

The analog is not subtle. Pre-2008 credit default swaps on senior tranches of subprime mortgage bonds also paid small premiums, because their issuers (and buyers) assumed that the defaults were uncorrelated. They were not uncorrelated. The correlation manifested all at once, and the small premiums did not come remotely close to compensating the writers of the protection.

What a cascade looks like

The Nexus Mutual payout I covered last week — $1.4 million across 62 restakers — was a good small test case, and it validated one specific insurance design. It did not validate the overall system's ability to absorb a correlated slash across, say, the top three AVSes simultaneously. That event would slash on the order of $2-4 billion of restaked capital. The insurance pool sized to cover it does not exist.

The uncomfortable conclusion

I have no strong view on which AVS will cause the cascade. I have a strong view that the current configuration of the industry — where restakers chase yield across an increasingly thinly capitalized set of services, and where the correlation risk is unpriced — is the kind of setup that historically produces a bad afternoon for the participants. Not "if." When.

None of this makes restaking a bad idea. It makes the current implementation of restaking a mispriced one. The difference matters, and when the mispricing corrects, it will correct all at once. If you are restaking across multiple AVSes today, I would at minimum know the name of each AVS, the identity of its security committee, and the specific conditions under which your capital can be slashed. Most restakers cannot pass that test. That is the problem.