What happened: a Teraswitch routing bug and a Solana finality halt scare

Solana came within about 4.5 percentage points of a real finality halt on Wednesday, August 12, after a routing fault at hosting and network provider Teraswitch knocked out roughly 29% of all staked SOL within the same few minutes. SOL itself barely moved — it's trading around $75.90 as of Thursday morning UTC, up a flat 0.3-0.6% on the day — because no funds were ever at risk. This was a plumbing failure, not a hack, an exploit, or an exchange outage. But it's the closest the network has come to an actual halt since the multi-hour outage in February 2024, and it exposes a structural weakness that hasn't been fixed since then.

The root cause was a bad default route. Teraswitch's Miami site pushed out a stripped-metric route that edge routers in London, Amsterdam, Dublin, Frankfurt, Singapore and Tokyo misread as locally originated, which cut twelve sites off from the rest of the network simultaneously. That's a classic BGP misconfiguration — the internet's routing protocol occasionally propagates a bad path this way — and Teraswitch fixed it within roughly 10 to 33 minutes, depending on the site. Fast, as these things go. The problem is what that single provider represented.

Was Solana down, and is my SOL at risk?

No, and no — not from this incident specifically. Solana did not halt, transactions kept confirming, and nobody's staked or unstaked SOL was frozen, lost, or exposed to a third party. The outage was a validator-connectivity problem, not a smart-contract bug or a custody failure, so there's no action for a typical holder or staker to take. That's also the reason the market shrugged: price action responds to capital at risk, and none was.

The risk this incident surfaces is different and slower-moving: it's about the network's ability to keep confirming transactions at all if a similar event goes slightly worse next time. That's a protocol-health question, not a today's-balance question, and it's worth understanding because it recurs.

How close did Solana get to a real halt?

Solana halts finality if more than 33.34% of staked SOL goes offline or acts inconsistently at once — that's the threshold beyond which the network can no longer get the supermajority agreement it needs to finalize blocks. During Wednesday's incident, delinquent stake peaked at roughly 28.83%, which put the network about 4.5 percentage points, or roughly 86% of the way to that line. It's the narrowest margin since the February 2024 halt, though it did not cross it.

The reason it got that close comes down to concentration, not bad luck. One autonomous system — AS20326, the internet routing entity tied to Teraswitch's infrastructure — hosts 27.34% of all staked SOL, equivalent to about 118.9 million SOL. That's already above the Solana Foundation's own delegation-program cap of 25% per network. When roughly 94% of that AS went dark in the same window as the routing fault, it moved the needle almost the whole way to the halt threshold by itself. A second factor made it worse: 59 validators holding about 80.2 million SOL stayed offline through the network's reconvergence instead of failing over to backup routes, among them Helius, Solana's second-largest validator, which was down for the full incident window.

Why one network alone could nearly break finality

This is the part of the story that outlasts Wednesday's headlines. Solana's finality model assumes stake is spread across enough independent infrastructure that no single failure — a data center outage, a routing bug, a cloud provider incident — can take out a third of the network at once. AS20326 sitting above the Foundation's own 25% cap means that assumption was already broken before Teraswitch's router hiccuped. The Foundation set that cap presumably because it understood the risk; the fact that one AS exceeded it anyway suggests either the cap isn't being enforced against delegated stake, or delegation flows faster than anyone is reviewing where it lands.

Marinade Finance, a major Solana liquid-staking protocol, published a post-mortem acknowledging the concentration problem directly and noting its own delegation model has a separate AS carrying an even higher 36.94% share. That's an unusually candid admission from a protocol with a direct stake in appearing decentralized, and it suggests the industry response is likely to be a real review of per-network and per-datacenter limits rather than a quiet dismissal.

What happens next

The base case is that this stays a decentralization story for validators and stakers, not a price catalyst. Marinade has already committed to reviewing concentration limits and publishing failover metrics, and if that follows through — stake genuinely diversifying away from AS20326, and validators demonstrating they can hot-swap onto backup routes instead of sitting offline through reconvergence — Wednesday reads in hindsight as the guardrail working: the network got to 86% of the threshold and stopped there.

The risk case is that nothing structurally changes. AS20326 stays above the cap, other autonomous systems carry similar hidden concentration the way Marinade's own delegation does, and the next single-provider fault — at Teraswitch or elsewhere — starts from a worse position and actually crosses 33.34%. That's the scenario that produces a repeat of February 2024's multi-hour halt, and it's the one that would finally move price, since a real halt freezes transactions network-wide rather than just rerouting traffic around a bad BGP announcement. For now, watch whether Marinade's review and any Solana Foundation enforcement of its 25% cap actually ship, rather than the SOL chart — that's where this story's next real update will come from.

Sources