Solana came close to losing transaction finality on Wednesday morning after a routing failure at hosting provider Teraswitch took 28.83% of staked SOL offline. The network requires roughly two-thirds of stake to remain active to finalize blocks. Once more than 33.34% goes dark, finality stops even if blocks continue to be produced. The incident left the chain 4.51 percentage points short of that limit, or about 86% of the distance to a halt in irreversible confirmations.

What Triggered the Outage

Teraswitch published a technical explanation of the event. The company uses an internal default route to indicate that an edge router can reach the wider internet. Each site normally prefers the route originated by its own routers. A default route originating from the Miami site was propagated with its metric and communities stripped. A route reflector in Amsterdam then pushed the faulty route into Europe and Asia-Pacific regions.

Edge routers in those locations treated the route as locally originated and preferred it over valid alternatives. When the invalid route reached data-center cores it was rejected, leaving twelve sites in London, Amsterdam, Dublin, Frankfurt, Singapore and Tokyo without a usable path. North American facilities remained unaffected. Engineers identified the problem within ten minutes. Service was restored at 04:16:15 UTC.

Concentration of Stake Exposed

Analysis by staking protocol Marinade Finance showed the outage was heavily concentrated on one autonomous system. AS20326 carries 118,890,767 SOL, more than a quarter of all staked tokens on the network. Ninety-four percent of that stake went offline at the same time. The Solana Foundation’s delegation program sets a 25% cap on any single autonomous system; AS20326 already stood at 27.34%.

Roughly 90 validators were affected. Fifty-nine of them, controlling 80.2 million SOL, returned online in a narrow window across Amsterdam, Frankfurt and Tokyo after waiting for routing tables to reconverge rather than failing over to alternative paths. Helius, one of the largest validators, remained offline for the full 33 minutes of the incident. Of 74 operators measured, only three recovered cleanly without interruption: Laine, Cogent Crypto and Lion3d.

An additional 14.1 million SOL went offline in the same period across other providers including latitude.sh, Limestone, Butterfly Research and Allnodes. That broader impact suggests that measuring concentration solely by hosting provider understates the degree of correlated risk.

Impact on Validators and Delegators

Validators that went offline missed staking rewards. The combined shortfall totaled 333 SOL, which will be covered by validator bonds at the end of the current epoch. Solana does not slash stake for downtime; operators simply forgo rewards while offline. Those missed rewards flow through to delegators as lower returns for the period.

Had the offline stake crossed the one-third threshold, no transactions would have finalized anywhere on the network. Bonds do not cover that scenario. The last time Solana experienced a full halt, in February 2024, recovery took nearly five hours.

Network Resilience and Upcoming Changes

Before Wednesday’s event, Solana’s official status page had shown continuous operation for more than 30 months. That record tracks full network stops rather than periods in which blocks continue to be produced while finality is impaired. Reports earlier in 2026 recorded as many as 32 individual validator delinquencies in a 30-day window, most caused by hardware, software or connectivity issues. The scale of simultaneous failure on Wednesday pointed to a systemic rather than isolated problem.

Solana is developing the Alpenglow protocol upgrade, which targets finality times of roughly 100 to 150 milliseconds. The design prioritizes safety over continuous liveness: the network would pause block production rather than risk confirming inconsistent transactions. Alpenglow also distinguishes between actively malicious validators and those offline because of passive failures such as the routing incident.

Marinade Finance examined its own stake distribution after the event and reported that four autonomous systems hold two-thirds of the stake allocated by its model, with one accounting for 36.94%. The firm stated it would review concentration limits by network and data center and begin publishing whether validators maintain hot-swap capacity and automatic failover.

The episode tested the boundary of Solana’s fault-tolerance assumptions more sharply than any event since the February 2024 outage, while leaving the chain short of an actual loss of finality.