TeraSwitch outage nearly halted Solana: 28.83% of the stake went offline
On August 12, the Solana network came dangerously close to a complete halt in transaction finalization. The cause was a routing failure at the infrastructure provider TeraSwitch, which simultaneously took down validators controlling 28.83% of all staked SOL. This was just 4.5 percentage points below the critical threshold of 33.34%, after which the blockchain stops confirming operations.
The incident affected about 90 validators operating through the autonomous system AS20326. This infrastructure accounted for 118.9 million SOL, or 27.34% of the network's total stake. At the moment of the failure, 94% of that amount went offline—an unprecedented scale of risk concentration.
Fortunately, a catastrophe was avoided. Routing was restored in about 33 minutes, and the validators returned to the network. However, during the downtime, they missed out on approximately 333 SOL in rewards—a symbolic amount, but the signal itself is alarming.
Infrastructure Concentration—A Hidden Threat
This case is a vivid illustration of a long-standing Solana problem: excessive dependence on a limited number of hosting providers. Even before the incident, as of July 22, TeraSwitch served validators with 27.1% of staked tokens. It was followed by UAB Cherry Servers (12.7%) and Latitude.sh (11%). Such concentration makes the network vulnerable to single technical failures.
Some major operators have already recognized the risks. For example, Coinbase distributes its 13 validators across TeraSwitch and 10 across Latitude, while using a backup server in a different location for each. This is a sensible approach, but it does not solve the problem systemically.
It is worth recalling that in November 2022, German hosting provider Hetzner already disconnected servers running Solana nodes—at that time, about 40% of validators with 20% of the stake were affected, and the network held up. But the situation now is different: the number of active validators has dropped to 800—a low not seen since 2021—whereas at the peak in 2023, there were more than 2,500.
My conclusion: Solana has once again demonstrated resilience, but this is more luck than systemic reliability. Until the community addresses the issue of infrastructure decentralization, every similar failure will serve as a reminder that a single provider can jeopardize the entire network. I recommend that validators and delegators reconsider their distribution strategies across autonomous systems—and the sooner, the better.