---
title: "Solana Came Close to a Finality Halt When 28.83% of Stake Went Offline"
description: "A routing fault at Teraswitch's Miami site knocked 90 Solana validators and about 125 million SOL offline for 33 minutes, pushing delinquent stake to 28.83% against the 33.34% level where the network stops finalizing blocks. Marinade Finance's incident analysis found more than a quarter of all staked SOL concentrated on a single ASN, with 94% of it going dark at once."
author: "Mike Hale"
published: 2026-08-12T14:04:00.000Z
updated: 2026-09-01T19:37:42.235Z
category: "Ecosystem"
tags: ["Solana"]
canonical: https://stacksmith.xyz/articles/solana-teraswitch-outage
source: "Stacksmith"
---

# Solana Came Close to a Finality Halt When 28.83% of Stake Went Offline

A routing fault at hosting provider Teraswitch took roughly 90 Solana validators and about 125 million SOL offline for 33 minutes on August 12, 2026, pushing delinquent stake to 28.83% against the 33.34% level where Solana stops finalizing blocks. The network never halted and users saw no disruption. It came within 4.51 percentage points, roughly 20 million SOL, of losing normal finality.

Blocks kept flowing because 597 of Solana's 699 validators stayed online. The affected validators missed roughly 333 SOL in rewards, a number that will be covered by validator bonds at the end of the epoch.

Most of what's known about the incident comes from [Marinade Finance's analysis](https://x.com/MarinadeFinance/status/2087485897470083337), which mapped stake to ASNs, timed each validator's recovery, and published the numbers within hours.

![Solana got 86% of the way to a halt — Marinade Finance chart](https://images.ctfassets.net/w6uv47mkdn6b/5LWsyHn7PR74QXDT8gyjke/db1b55fafbc764112981cbea3839c25c/marinade-solana-teraswitch.jpg)

*Image courtesy Marinade Finance*

## What actually broke inside Teraswitch's network

A misconfigured default route out of Teraswitch's Miami site spread to its sites across Europe and Asia-Pacific, and routers there started sending traffic down a path that didn't work. Twelve sites, spanning London, Amsterdam, Dublin, Frankfurt, Singapore, and Tokyo, lost their connection to the rest of the network. Validators at those sites went unreachable, stopped voting, and their stake went delinquent. North America was untouched.

Teraswitch found the bad route in about ten minutes and restored service by 04:16:15 UTC, a 33-minute outage in total. There was no exploit, no attacker, nothing to trace on Solscan. This was a networking failure. What matters is how close the rest of Solana came to losing finality while those validators were dark.

## How close did Solana come to losing finality?

Solana needs more than 66.67% of stake actively participating in consensus to finalize blocks normally, which puts the failure point at roughly 33.34% delinquent stake. The peak was 28.83%, or about 86% of the way there.

Losing finality would not have meant stolen funds, a corrupted ledger, or validators shutting off. It would have meant new blocks stop becoming irreversible until enough voting stake returns. That's a serious operational problem for exchanges, bridges, custodians, and anything that settles against finalized blocks. Solana's status page is right not to call this a mainnet outage, but calling it a non-event would be just as wrong.

A real finality stall isn't a quick fix. The [February 2024 halt](https://solana.com/news/02-06-24-solana-mainnet-beta-outage-report) took roughly five hours to restart the cluster.

## One ASN holds more than a quarter of staked SOL

Marinade's analysis put 118,890,767 SOL on Teraswitch AS20326, an autonomous system number (ASN), the block of internet address space one network operator controls and routes traffic for. That ASN alone holds about 27.34% of all staked SOL, and roughly 94% of it went dark at once.

A network can have a high validator count and still carry systemic risk if a lot of those validators sit behind the same ASN, hosting provider, routing layer, or upstream carrier. Solana's superminority, the smallest group of validators that could stop finality by going offline together, is 17 validators.

The Solana Foundation already builds this concern into its delegation program: participating validators must run on an ASN and hosting provider holding less than 25% of total network stake, with a separate 15% datacenter limit. Those are delegation requirements, not protocol-enforced caps, but this incident is exactly why they exist.

Marinade also saw another 14.1 million SOL go offline in the same minutes on other providers, including validators attributed to Latitude.sh, Limestone, Butterfly Research, and Allnodes. There's no public evidence Teraswitch caused those outages, and the data doesn't say whether it was a shared dependency or coincidence. Either way, counting stake by hosting provider undercounts what actually fails together. Two validators can sit with completely different companies and still share a transit provider, a routing dependency, or a physical facility further down the stack.

## Why didn't validator failover kick in?

Marinade found 59 validators holding 80.2 million SOL returning inside the same narrow window, with matching patterns in Amsterdam, Frankfurt, and Tokyo. They came back when the routing reconverged, not before.

Of the 74 validators Marinade could evaluate, three appeared to fail over cleanly: Laine and Cogent Crypto, both operated by Sol Strategies, plus Lion3d. Helius, one of the largest operators on Solana, was reportedly down for the full 33 minutes.

That doesn't prove the rest had no backup infrastructure. It does show that whatever redundancy existed didn't get them back into consensus before the shared network recovered on its own. From the outside, you can't tell whether an operator has no second site or a failover that never triggered, and that's its own problem for anyone choosing where to stake.

## What changes after the near-miss

Marinade turned the same analysis on itself and published that too. Four ASNs hold two thirds of the stake its SAM allocation system distributes, and AS395201 alone accounts for 36.94%.

> "Nobody should be comfortable with that, us included." - *Marinade Finance*

Marinade has two changes queued: reviewing concentration limits per ASN and per datacenter, and publishing whether a validator runs hot spare infrastructure and automatic failover, something that's currently invisible from outside.

If you run a validator, the questions to answer after the Solana Teraswitch outage are which ASN you announce from, which upstream carriers sit behind it, whether your failover has ever been tested against a routing failure and not just a dead machine, and whether your backup site shares any of those dependencies with your primary.

Despite what happened, Solana held up, with blocks finalizing the entire time, on the strength of the hundreds of validators spread across other networks and regions that never went dark. But 28.83% delinquent stake from a single hosting provider's mistake is just 4.51 percentage points from a threshold that would have stalled finality network-wide. The resilience that carried Solana through this incident is exactly the case for spreading stake across more ASNs, more datacenters, and more regions.