A single malformed BGP route originating from Teraswitch Networks' Miami facility on August 12 knocked 28.83% of all staked SOL offline for 33 minutes, pushing the Solana network to 86% of the threshold at which transaction finality ceases. One autonomous system — AS20326, operated by Teraswitch ...
"If delinquency had gone past a third, nothing finalizes for anyone holding SOL anywhere, and there's no bond for that. The February 2024 halt took about five hours to restart." — Marinade Finance
A single malformed BGP route originating from Teraswitch Networks' Miami facility on August 12 knocked 28.83% of all staked SOL offline for 33 minutes, pushing the Solana network to 86% of the threshold at which transaction finality ceases. One autonomous system — AS20326, operated by Teraswitch — held 118.9 million SOL, more than one-quarter of total network stake. Roughly 94% of that stake went delinquent simultaneously.
The network continued producing blocks and processing transactions throughout the incident. No funds were lost. But the episode exposed a structural vulnerability that protocol-level code cannot fix: the physical infrastructure layer beneath Solana's validator set is concentrated enough that a routine configuration error at a single hosting provider can bring finality within 4.5 percentage points of failure.
This report examines the incident mechanics, the infrastructure concentration dynamics that enabled it, and the broader implications for proof-of-stake networks that depend on commercial hosting providers for validator uptime.
At approximately 03:43 UTC on August 12, 2026, validators hosted on Teraswitch Networks infrastructure began losing connectivity. Within minutes, 102 of 699 staked validators stopped voting. By headcount, that represented roughly 15% of validators. By stake weight, the damage was far worse: 28.83% of all staked SOL went delinquent, according to data published by Marinade Finance.
Solana's consensus mechanism requires participation from validators representing at least 66.67% of staked tokens to finalize transactions. The corollary: if more than 33.34% of stake goes offline, finality halts entirely. The August 12 incident left approximately 20 million SOL — roughly 4.5 percentage points — of headroom before that threshold would have been breached.
The 597 validators that remained online continued producing blocks and landing transactions. Jacob Creech, a technology executive at the Solana Foundation, confirmed that "blocks kept producing and transactions kept landing" throughout the disruption. Teraswitch engineers identified the routing fault within 10 minutes. Traffic was restored by 04:16:15 UTC. Total elapsed time from first delinquency to full restoration: approximately 33 minutes.
Affected validators collectively lost an estimated 333 SOL in staking rewards during the downtime window.
The failure originated at Teraswitch's MIA1 facility in Miami. A routine configuration change introduced a malformed default route that was advertised without proper BGP attributes. Under normal conditions, such a route would be filtered at the edge. Instead, it reached a route reflector at Teraswitch's AMS2 facility in Amsterdam.
The Amsterdam reflector propagated the altered route into European and Asia-Pacific markets. Local routers at 12 sites — London, Amsterdam, Dublin, Frankfurt, Singapore, and Tokyo — preferred the malformed route over valid forwarding paths. The result: simultaneous loss of reachability across two continents.
North American sites were unaffected. The malformed route did not propagate into domestic U.S. peering fabric, which is why the geographic impact was asymmetric.
Teraswitch subsequently stated that the configuration change that triggered the incident was removed, and additional protections and software upgrades were applied. The company did not disclose specifics of the misconfiguration or explain why its route reflector lacked adequate filtering.
The incident's severity was a direct function of infrastructure concentration. AS20326, the autonomous system number registered to Teraswitch Networks Inc. (Amsterdam-based), hosted validators representing 118.9 million SOL — over 26% of total staked supply. When the routing failure hit, 94% of that stake went offline in a single event.
Beyond Teraswitch, an additional 14.1 million SOL went delinquent across validators hosted on Latitude.sh, Limestone Networks, Butterfly Research, and Allnodes infrastructure. These secondary outages suggest shared upstream dependencies or peering relationships that amplified the blast radius.
Marinade Finance's post-incident analysis revealed further concentration within its own delegation portfolio: four ASNs account for two-thirds of Marinade-allocated stake, with AS395201 alone representing 36.94% of Marinade's delegation.
The concentration problem is not new. Solana's validator infrastructure has historically clustered around a small number of hosting providers. At various points in 2023 and 2024, Hetzner and OVH individually held well above 25% of network stake before Foundation policy changes forced redistribution.
Perhaps more concerning than the concentration itself was the near-total absence of automatic failover. Marinade Finance's analysis of 74 affected validators found that only three returned through automatic failover systems — operators running Sol Strategies and Lion3d infrastructure. The remaining 71 validators waited for Teraswitch to restore connectivity.
Helius, Solana's second-largest validator by stake delegation, remained offline for the entire 33-minute duration. Fifty-nine validators holding a combined 80.2 million SOL stayed delinquent until the Teraswitch network itself recovered.
This failover data indicates that the vast majority of Solana validators, including the largest operators, lack multi-provider redundancy. Running a backup node on an alternative hosting provider is technically feasible but economically unattractive given Solana's hardware requirements, which represent monthly operating costs of $1,400 to $3,400 or more per validator. Maintaining a hot standby on a separate ASN effectively doubles infrastructure costs.
The concentration problem compounds against a backdrop of declining validator counts. Solana's active validator set peaked at approximately 2,560 in March 2023. By late January 2026, the count had fallen to 795 — a 68% decline. As of August 13, 2026, the seven-day moving average stood at 691 validators, continuing a steady drop from more than 750 in May.
The decline is driven by economics. The Solana Foundation reduced subsidies for voting costs and staking matching policies beginning in April 2025 under a "pruning" initiative that systematically removed underperforming validators. Monthly operating costs for a small-scale validator range from $1,400 to $3,400 or more, while validators with only a few thousand SOL staked earn rewards insufficient to cover baseline expenses.
The paradox, as noted by multiple analysts, is that the surviving validator set may be more operationally capable than the larger set it replaced. Client diversity has improved: Firedancer (or its hybrid variant, Frankendancer) ran on more than 20% of active validators by Q2 2026, up from zero before its December 2025 mainnet launch. The Foundation's delegation program now spans 59 delegations across 24 countries.
But fewer validators mean each remaining node carries more stake weight, which amplifies the systemic impact when a shared infrastructure provider fails. The August 12 incident demonstrated this arithmetic directly.
The Solana Foundation Delegation Program implemented concentration caps effective May 1, 2026:
These caps applied before the August 12 incident. Teraswitch's AS20326, holding 26% of staked SOL, was already above the 25% ASN threshold — but the caps apply only to Foundation-delegated validators, not to the network at large. Validators receiving stake from private delegators or liquid staking protocols face no such restrictions.
Marinade Finance stated after the incident that it plans to review its own concentration limits. No specific policy changes have been announced.
The Solana Foundation's official status page recorded no mainnet incident for August 12. Cluster uptime was reported at 100% over the 90-day period. This is technically accurate — the network did not halt — but it obscures the proximity to failure.
The Teraswitch incident is not unique to Solana in concept, though the network's architecture makes it particularly susceptible. Solana's hardware requirements — high-throughput CPUs, 512 GB+ RAM, NVMe storage, 10 Gbps networking — constrain validators to commercial data centers. Unlike Ethereum, where validators can run on consumer hardware, Solana's performance-oriented design creates structural dependency on professional hosting infrastructure.
Ethereum's validator set of approximately 1 million validators (operated by a smaller number of distinct entities) distributes risk differently. A single hosting provider outage on Ethereum would affect a smaller fraction of total stake simply because the validator count is orders of magnitude larger and hardware requirements are lower.
However, Ethereum faces its own concentration risks. According to an OpenZeppelin technical risk assessment published in 2026, multiple blockchain networks show elevated hosting provider concentration when measured by stake weight rather than node count.
The fundamental tension is between performance and resilience. High-performance chains require expensive hardware, which concentrates validators in professional data centers, which creates correlated failure modes. Lower-performance chains distribute more broadly but sacrifice throughput.
For institutional users evaluating Solana for settlement or custody applications, the August 12 incident quantifies a previously theoretical risk: a 33-minute near-miss at 86% of finality failure, caused by a routine ISP configuration error.
Solana's consensus mechanism performed as designed on August 12. The protocol did not fail. But the infrastructure layer beneath it demonstrated that a network processing billions of dollars in daily volume can be pushed to within 4.5 percentage points of finality failure by a misconfigured route at a single hosting provider.
The 33-minute incident produced no lasting damage: no funds were lost, no blocks were missed by the supermajority, and recovery was swift. But the structural conditions that enabled the near-miss — a shrinking validator set, high hardware barriers to entry, and commercial hosting concentration — remain unchanged. Concentration caps apply selectively. Failover infrastructure is absent at most validators. And the economics of running a Solana validator continue to push smaller operators out of the network.
The February 2024 full halt took approximately five hours to restart. The August 2026 near-miss lasted 33 minutes. The difference between the two was 20 million SOL of stake weight — roughly $3 billion at current prices — remaining online instead of going delinquent. That margin is not a safety feature. It is a measurement of how close the network came to an outcome its architecture could not prevent.