A routing failure at hosting provider TeraSwitch knocked validators representing 28.83% of the SOL securing Solana offline, according to an analysis published by staking platform Marinade on August 12. The network continued to finalise transactions, and its public status page recorded no outage.

The episode still matters because Solana stops finalising when more than one third of stake no longer votes. The gap between 28.83% and that threshold was narrow, while public data still places nearly 28% of active stake on one internet network. The risk lies less in the number of validators than in the infrastructure dependency shared behind them.

What happened

In a thread posted at 10:26 UTC, Marinade said 90 validators became delinquent, meaning they temporarily stopped voting with the rest of the cluster. Together they represented 28.83% of staked SOL. Marinade placed the loss-of-finality threshold at 33.34%.

The company traced the event to a bad default route advertised from a TeraSwitch facility in Miami. A route reflector in Amsterdam allegedly propagated it to European and Asian sites. Edge routers preferred the unusable route, leaving twelve sites without a valid path for forwarding traffic.

Marinade said traffic returned at 04:16:15 UTC, about ten minutes after the problem was identified. Some validators nevertheless remained offline for 33 minutes. The analysis put their missed rewards at 333 SOL and said validator bonds should cover them at the end of the epoch.

CoinDesk reported the episode later on August 12. Its article makes the event easier to follow, but the figures and technical account largely come from Marinade; it is not a second independent measurement of the routing failure.

A near miss, not a Solana outage

The central distinction is straightforward: validators lost connectivity, but Solana did not cross the threshold that would stop finality. The official status page showed all systems operational and no downtime for the day.

That does not contradict Marinade’s analysis. A status page reflects whether the network is delivering its public service; it need not flag every loss of internal redundancy while transactions keep finalising. This incident was closer to a deeply eroded safety margin than an outage visible to every user.

The one-third threshold comes from Byzantine fault-tolerant consensus. As long as more than two thirds of voting weight remains available and converges, the network can keep confirming one shared state. Once more than one third stops participating, the safer outcome is to stop finalising rather than risk incompatible histories.

GatherHub’s Blockchain course explains the difference between block production, validation and consensus. Here, the critical constraint was not Solana’s theoretical throughput. It was whether enough stake-weighted validators stayed connected at the same time.

Validator count is not enough

A network can have many operators and still retain a common failure point. Two validators run by separate organisations are not fully independent if they rely on the same host, autonomous internet system, power supply or failover path.

Validators.app data, checked on August 12, associated TeraSwitch’s AS20326 with 87 validators and 121.58 million active SOL, or 27.96% of stake. None of those validators was still marked delinquent when checked. The figures differ slightly from Marinade’s 118.89 million SOL and 27.34% because stake and rankings move over time.

That concentration exceeds a benchmark published by the Solana Foundation. Its delegation criteria have required participants since May 1, 2026 to operate on an ASN or hosting provider holding less than 25% of overall network stake. They also set a 15% limit for a single data centre.

These are not consensus rules automatically enforced across every validator. They determine eligibility for stake delegated through the Foundation’s programme. They nevertheless recognise the underlying problem: internet-provider concentration can create systemic risk even when servers are spread across several cities.

Why it matters now

For users, the episode caused no measured public disruption. Transactions continued and the status page logged no downtime. Calling it a “Solana outage” would therefore be inaccurate.

For validators and delegators, the impact was more tangible: missed votes, reduced rewards and evidence that backup arrangements did not recover quickly enough. Marinade said it observed a clean recovery at only three of the 74 operators it could measure. It also acknowledged that external data cannot distinguish a missing secondary site from a failover that was present but misconfigured.

For applications, stablecoins, exchanges and financial protocols built on Solana, the risk is collective. Had offline stake crossed one third, finality would have stopped network-wide, including for services hosted elsewhere. Software can be distributed while the physical and internet layers beneath it remain much more concentrated.

GatherHub’s inference is that useful decentralisation must be assessed by failure domain: operator, ASN, data centre, region, power and validator client. Counting nodes without mapping these dependencies overstates practical resilience.

Limits and open questions

The main technical narrative available at publication came from Marinade, a company directly involved in staking quality and allocation. TeraSwitch had reportedly published a write-up, but its status site was not accessible from our verification environment. GatherHub therefore treats the routing details as Marinade’s account, not as an independent provider report.

Marinade also identified another 14.1 million SOL that went offline at other providers during the same window. It said it could not determine whether there was a shared dependency or a coincidence. This article does not attribute that additional stake to TeraSwitch.

Validators.app describes declared ASN placement and active stake at a point in time. It cannot expose every subcontracting relationship, physical link or failover procedure. Conversely, separate sites on one ASN may still have different local protections.

What to watch

The first useful piece of evidence would be an accessible TeraSwitch post-mortem identifying the root cause, geographic scope, failed safeguards and measures designed to prevent another route propagation event.

The second is AS20326’s share of stake. A move below 25% would reduce the weight of one network failure, although it would not by itself guarantee diversity across facilities and operators.

The third is failover. Marinade said it will review concentration limits by ASN and data centre and publish whether validators use hot swap and automatic failover. Observable tests will matter more than simply listing a backup site.

Solana did not go down on August 12. That is precisely what makes the incident useful: it reveals a weakness before it becomes a visible outage and provides a concrete threshold against which corrective action can be measured.

Sources consulted

Editorial illustration generated by GatherHub Editorial under the “Portail Matière” identity. This article is not financial advice.