A single misconfigured default route out of Teraswitch's Miami site spread across its European and Asia-Pacific network; of 74 measured operators, only three recovered cleanly.
A routing fault inside Teraswitch's network pushed enough staked SOL into delinquency early on Wednesday, August 12, to leave Solana roughly 4.5 percentage points below the 33.34% mark at which the network stops finalizing transactions. A validator is delinquent when it abruptly falls out of the set of nodes taking part in consensus; finality is the point at which transactions can no longer be reversed.
The trouble began with a default route tied to Teraswitch's Miami site, advertised after its metric and its routing communities had been stripped off. A route reflector at the company's Amsterdam AMS2 site pushed the altered route out across its European and Asia-Pacific markets, where edge routers preferred it to legitimate local routes. Core routers further down the chain treated it as invalid and refused it, leaving twelve sites with no usable default path onto the internet and no reachability to one another across the backbone. Teraswitch's remaining North American sites came through unaffected.
One provider's routing fault reached that much stake because AS20326, the autonomous system PeeringDB identifies Teraswitch as operating, held roughly 118.89 million SOL — more than a quarter of all staked SOL — and 94% of it dropped offline at the same moment. A further 14.1 million SOL, spread across four other providers, went delinquent in those same few minutes; Marinade, which hands out staked SOL to validators through its own allocation model, said its data left it unable to tell whether that pointed to a dependency they had in common or was simply chance.
Teraswitch said its engineers pinpointed the malformed route inside 10 minutes of it starting and pulled MIA1 off the backbone so it could spread no further. The sites that had been hit converged back onto their local default routes, with service restored at 04:16:15 UTC. The company said it had not yet identified whether the underlying defect lay in the Amsterdam route reflector or in the Miami edge routers, and that a full root-cause report would follow.
The realised damage was small. Roughly 90 validators missed a combined 333 SOL in rewards, a loss validator bonds will absorb, so stakers bear none of it. Marinade's concern is the other side of that line: had delinquency gone past one-third of stake, nothing would have finalized for anyone holding SOL anywhere, and no bond covers a loss of that kind. In February 2024, a software bug stalled more than 95% of cluster stake and block finalization stopped for roughly five hours.
Operator readiness was the second exposure. Of the 74 validator operators Marinade had measurements for, three came back cleanly. Helius, an infrastructure provider and Solana's second-largest validator, stayed offline for the full 33-minute disruption. And 59 validators, holding 80.2 million SOL between them, all returned inside one tight span, having waited for routing to converge again instead of switching over to any backup.
Marinade said it will re-examine the ceilings it sets on how much stake may pile up in a single autonomous system or a single data center, and will start making public which individual validators run hot-swap machines and automatic failover. It turned the same lens on itself: two-thirds of what its own allocation model distributes goes to just four autonomous systems, with AS395201 alone accounting for 36.94%, and Marinade said no one — itself included — should be comfortable with that.
Marinade said Solana had come 86% of the way toward a halt that morning and that the episode drew almost no notice anywhere; Solana's official public status page listed no incident for August 12.