Routing error knocked 29% of Solana stake offline

On Aug. 12 a routing error at Teraswitch cut off validators that controlled nearly 29% of Solana stake, raising skipped slots above 32% and reducing non-vote transaction throughput to under 300 transactions per second. Blocks continued to be produced, but finality was delayed for about 30 minutes.

Teraswitch traced the failure to a stale default route on a Miami edge router. Routine maintenance at a transit provider activated that route, and transposed values in a routing policy caused a no-export instruction to apply toward Europe and the Asia-Pacific region. The invalid route spread through Teraswitch’s network and prevented affected data centers from reaching the internet via healthy local edge routers.

Twelve sites across Europe and the Asia-Pacific region lost internet and inter-site connectivity, including data centers hosting Solana validator and RPC nodes. Teraswitch’s provider traffic fully recovered 33 minutes after the first alarm, according to its postmortem.

Independent telemetry recorded skipped slots above 32% and non-vote throughput falling below 300 transactions per second, down from a typical range near 1,100 to 1,300 tps. Observers measured about 30 minutes during which newly produced blocks had not reached finality. When connectivity returned, the backlog of transactions finalized.

The Solana Foundation reported that the affected validators stopped voting and that the offline stake was roughly 29%, about four to five percentage points below the 33% threshold that would halt finality. Because validators continued to produce blocks, the network did not experience a chain-wide production stop.

The Foundation previously reported that Teraswitch hosted 38% of network stake in 2025 and that hosting had been reduced to below 30% before the outage. Many separately operated validators used the same provider, so a single infrastructure failure affected a large group of validators at once.

Teraswitch applied provider-side hardening on the day of the outage to prevent an invalid route from stopping sites from using local edge routers. Its postmortem noted that additional route monitoring was being implemented.