- 28.83% of all staked SOL stopped voting early Wednesday, according to staking Marinade Finance. At 33.3%, Solana stops finalising blocks.
- Around 90 were affected, and they gave up 333 SOL in rewards between them.
- The chain kept processing transactions the whole time. It was a near miss rather than an outage, and near misses are the cheapest way to learn anything.
Solana did not go down on Wednesday. It came close to something quieter and, in some ways, more awkward: a state where blocks keep being produced but the network can no longer declare any of them permanent. Marinade Finance put the gap at roughly 4.5 percentage points.
Finality is the moment a payment stops being provisional and becomes permanent. If a loses that, it doesn’t stop. It carries on taking transactions while nobody can say for certain that any of them will stick, and the usual response from exchanges and payment firms is to freeze deposits and withdrawals until it comes back.
So the practical version for anyone holding SOL or using an app built on it: your money is not stolen and nothing is broken, but you may not be able to move it, and the screen won’t necessarily tell you why.
What “delinquent” means here
Solana is run by validators: independently operated computers, several thousand of them, that keep a copy of the ledger and agree on what happened. They don’t get an equal say. Influence is weighted by stake, meaning the amount of SOL that has been locked up and assigned to each one, either by the operator or by ordinary holders delegating their coins.
Validators vote on blocks continuously. A validator is marked delinquent when it stops sending those votes, usually because it has crashed, fallen behind the rest of the network, or lost connectivity. It hasn’t left, and it hasn’t done anything malicious. It has simply gone silent.
Marinade’s count covers stake, not machines. So 28.83% delinquent doesn’t mean 28.83% of validators fell over. It means the validators that went quiet controlled nearly three-tenths of the network’s voting weight between them.
Why two-thirds is the number that matters
Solana finalises a block once validators representing more than two-thirds of all staked SOL have voted on it. That threshold isn’t arbitrary. It’s the standard design for this kind of consensus, and it exists so the network can survive up to a third of its participants behaving badly or vanishing without ever confirming two contradictory versions of history.
The arithmetic runs the other way too. If more than one third of stake stops voting, the remaining two-thirds no longer exists, and nothing can be finalised. That is the 33.3% line. At 28.83%, Solana was inside its own safety margin, but not by much.
Had it crossed, the chain would have kept producing blocks. They just would have sat there unconfirmed, piling up, until enough validators came back to clear the backlog.
The bit that should bother people is the 90
Ninety validators dropping out inside the same window is not ninety independent accidents. Machines fail all the time, and the network is built to absorb that. What it is not built to absorb is a lot of them failing for the same reason at the same moment.
Correlated failure is the recurring weak point in every proof-of-stake network, not just this one. Validators that run the same client software hit the same bug together. Validators renting space from the same cloud provider go offline together when that provider has a bad afternoon. Validators in the same data centre lose connectivity together.
Marinade hasn’t set out a single cause, and until someone does, the honest position is that we know how much stake went quiet and how many operators it sat with, and not yet why. That “why” is the whole story. A one-off glitch is noise. A shared dependency that 90 operators didn’t know they had is a design problem that will happen again.
The financial damage, for context, was almost nothing: 333 SOL of missed staking rewards spread across 90 operators, which is a small sum each and no loss at all to ordinary holders. Nobody’s coins moved. The cost here is entirely in what it revealed.
Solana has form, and it also has history
This lands on a network with a reputation to manage. Solana suffered a series of full outages in 2021 and 2022, and a roughly five-hour halt in February 2024, each requiring validator operators to coordinate a restart by hand. The chain has run without a total stoppage since, and a lot of the engineering work in between was aimed squarely at making that streak hold.
Wednesday doesn’t break the streak. It does show the margin is thinner than the uninterrupted uptime suggests.
And the audience for that has changed. Nasdaq-listed treasury companies now hold SOL and book staking revenue from it, and Israel’s largest bank has just started offering solana trading to customers through Galaxy. Institutions do not price downtime the way traders do. They write it into risk committees.
What to watch
The post-mortem. Whether Solana’s core developers or the validator community publish what those 90 operators had in common, and whether it points to one client, one host or one region. That answer determines whether this was bad luck or a structural concentration nobody had measured.
Second, whether stake starts moving. The most direct fix is delegators spreading their SOL across operators running different software in different places. If the distribution of stake looks the same in three months, the same near miss is available whenever the same trigger comes round again.