On September 14, the Solana Foundation revisited a network event that occurred on August 12. A routing failure at TeraSwitch, a major infrastructure provider used by validators, temporarily knocked nearly 29% of the network's stake offline.

Blocks continued to be produced and transactions continued entering the chain. According to the Foundation, the infrastructure provider fully recovered in just over 30 minutes.

By the narrow definition of uptime based on block production, Solana never went down.

28.83% went offline against a 33.34% critical threshold

Contemporary network data put peak delinquent stake at 28.83%. Solana needs more than two-thirds of stake actively voting to finalize blocks.

Once roughly 33.34% becomes unavailable, the network can lose the supermajority required for finalization even if block production itself continues.

The gap during the incident was therefore only about 4.5 percentage points.

Solana Compass reported that roughly 90 validators were affected at the same time. The important figure was not the validator count itself but the amount of stake sitting behind a shared infrastructure dependency.

Blocks kept coming while finality became much less healthy

This is where different descriptions of the event diverge. The Solana Foundation emphasizes uninterrupted block production and therefore continued network uptime.

Independent monitoring from Metrika shows a more complicated picture. Its analysis found a roughly half-hour period in which blocks were still produced but did not reach observed finality, followed by a sharp clearing of the accumulated backlog as validators reconnected.

Metrika measured finalization delays reaching around 1,600 seconds during that recovery, compared with a baseline of roughly 13 seconds.

That is not the same as a complete blockchain halt. Blocks still existed. But for systems that treat finality as the boundary after which settlement is irreversible, the distinction is operationally important.

User transaction throughput dropped as well

The simultaneous validator loss affected other network metrics. Metrika observed skipped slots rising above 32%, compared with a normal baseline below 1%.

True user transaction throughput fell below 300 transactions per second at its trough, down from the roughly 1,100 to 1,300 TPS range visible in its normal observations.

The chain was not offline. It was also clearly not operating normally.

Infrastructure concentration was the real single point of failure

The Solana Foundation itself acknowledges the concentration problem. It says TeraSwitch hosted about 38% of network stake during 2025.

Work was then undertaken to reduce that share below 30%. The Foundation now points to that reduction as one reason the August 12 outage stayed below the consensus threshold that would have created a much more serious event.

The incident exposes a weakness in decentralization metrics that count validators without examining the infrastructure beneath them. Independent operators can still share the same hosting company, data center or network path.

They look separate until the route disappears.

Devnet lost far more stake and recovered without a coordinated restart

The Foundation also highlights what happened on Devnet. According to its account, roughly three-quarters of stake on the development network disappeared during the same infrastructure failure.

Devnet nevertheless recovered on its own without a coordinated restart. Solana uses that outcome as evidence that protocol and infrastructure changes made over several years have improved the network's ability to absorb and recover from certain failures.

Those changes include stake-weighted quality of service and a redesigned transaction scheduler developed after earlier periods of congestion and instability.

100% uptime can be true and still be incomplete

Solana has not experienced a full block-production outage since February 2024, allowing the Foundation to claim 100% uptime over that period.

The TeraSwitch incident shows why that number cannot describe network health on its own. A blockchain can continue producing blocks while finalization stalls, skip rates spike and user throughput collapses.

A casual transfer may still appear to work. A financial system that uses finality as its settlement boundary has a very different risk model.

On August 12, Solana did absorb the loss of 28.83% of its stake without a full network halt. Roughly another 4.5 percentage points would have put the event into a very different category.