TL;DR:
- A routing error at Teraswitch took 28.83% of Solana’s stake offline, coming close to the critical 33.34% threshold.
- Autonomous system AS20326 concentrates 27.34% of the network’s total stake, above the 25% limit set by the Solana Foundation.
- Around 90 validators were affected and collectively lost 333 SOL in rewards during the 33 minutes the incident lasted.
Solana was 4.51 percentage points away from losing the ability to finalize transactions on Wednesday morning. A configuration error in a default route at hosting provider Teraswitch took 28.83% of the network’s stake offline, when the threshold at which the chain stops finalizing blocks is 33.34%. The network reached 86% of the way toward a complete halt.
Teraswitch published a technical report explaining the origin of the failure. The company uses an internal default route to indicate that an edge router can reach the internet, and each site normally prioritizes the one originated by its own routers. A route from its Miami headquarters was propagated without its metric or its communities, and a route reflector in Amsterdam distributed it toward Europe and Asia-Pacific.
Routers in those regions interpreted it as locally originated, preferred it over the real one, and sent it to the data center core, which rejected it as invalid. Twelve sites in London, Amsterdam, Dublin, Frankfurt, Singapore, and Tokyo were left without valid routes. Engineers identified the fault within ten minutes and Solana’s service was restored at 04:16:15 UTC.
Solana and the Stake Concentration Problem
Staking protocol Marinade Finance analyzed post-incident data and found that the disruption was concentrated in a single autonomous system. AS20326 holds 118,890,767 SOL, more than 27% of the network’s total staked supply, and 94% of that volume went offline within the same minutes. That percentage exceeds the 25% limit per autonomous system established by the delegation program of the Solana Foundation.
Of the 74 operators Marinade was able to measure, 59 validators with 80.2 million SOL waited for routing reconvergence rather than migrating to another option. Helius, the network’s second-largest validator, was inactive for the full 33 minutes. Only three operators resumed activity without issues: Laine, Cogent Crypto —both managed by Sol Strategies— and Lion3d.
Marinade also applied the same analysis to its own allocation model and found that four autonomous systems concentrate two thirds of the stake it distributes, with one of them, AS395201, accumulating 36.94%.
The firm publicly acknowledged that it will revise its concentration limits by network and by data center, and that it will begin disclosing whether a validator operates with hot swap and automatic failover, information that is currently not verifiable from the outside. The 333 SOL tokens in lost rewards will be covered by validator bonds at the close of the epoch. Had the downtime exceeded one third of the stake, no transaction would have finalized for any holder, and no bond covers that scenario.
