Submarine cable health, March–April 2026: 16 high-severity events mapped
Between March 10 and May 2, 2026, GeoCables logged 82 latency anomalies across 49 submarine cables. Sixteen of those crossed our high-severity threshold — classified by our monitoring stack as events warranting immediate attention. The remaining 66 stayed in the warning band: visible degradations that resolved on their own.
This report walks through every classified event from those 53 days. The data shows three things at once: a handful of cables that destabilize repeatedly, several geographies where small-basin systems oscillate without breaking, and a global median resolution time of about 3.4 hours. Submarine cables, as it turns out, spend more time in the gray zone between “fine” and “cut” than most operators publicly acknowledge.
How we measure
Each cable in our catalog is associated with one or more landing-point pairs. For each pair, we run continuous round-trip-time measurements from probes positioned in or near the relevant landing cities — both RIPE Atlas anchors and our own probe network in Moscow, Saint Petersburg, Sevastopol, Almaty, Minsk, Tbilisi, and Jerusalem. Baseline RTT for each route is computed from a 30-day rolling window. When a current measurement deviates significantly from baseline, considering both magnitude and persistence, the system raises an alert. Two severity tiers exist: warning (deviation crosses the first threshold) and critical (deviation crosses the second, or persists across multiple measurement cycles).
Resolution is automatic. An alert moves to resolved after three consecutive measurement cycles return to baseline range. If no measurement returns within an extended window, the alert moves to monitoring — flagged as ongoing without auto-resolution.
The most-affected cables
Seven cables triggered three or more separate alerts during the 53-day window:
| Cable | Events | Of which high-severity | Max RTT seen (ms) | Baseline RTT (ms) | Affected segment |
|---|---|---|---|---|---|
| Jakarta–Bangka–Batam–Singapore (B2JS) | 8 | 2 | 116 | 43–117 | Tanah Merah, SG ↔ Jakarta, ID |
| Tannat | 5 | 4 | 232 | 82–249 | Las Toninas, AR → Santos, BR |
| COBRAcable | 4 | 1 | 45 | 26–34 | Eemshaven, NL ↔ Endrup, DK |
| Batam–Dumai–Melaka (BDM) | 4 | 0 | 61 | 30–165 | Batam, ID ↔ Melaka, MY |
| Ionian | 3 | 0 | 83 | 83–105 | Preveza, GR ↔ Crotone, IT |
| Asia–America Gateway (AAG) | 3 | 0 | 117 | 109–122 | Changi North, SG ↔ Morro Bay, US |
| Trapani–Kelibia 2 (KELTRA-2) | 3 | 0 | 84 | 43–72 | Kelibia, TN ↔ Trapani, IT |
Two of these stand out for different reasons. Tannat triggered four high-severity alerts in fourteen days — April 17, April 28, April 29, April 30 (twice on May 2) — all on the same Las Toninas, Argentina → Santos, Brazil leg. Five separate events on a single corridor in two weeks, four of them severe, is a different signal from one cable that drifted once. Something on this segment is genuinely unstable, whether at the cable level, at the peering layer immediately after the cable, or at the measurement endpoints.
B2JS is the opposite shape. Eight events — the most of any cable in the window — but only two crossed the high-severity threshold, and the cable predominantly serves Singapore ↔ Jakarta, a route with abundant alternative paths through Apricot, SJC, and the regional peering fabric. High frequency, low individual impact, fast resolution. The pattern reads as routine micro-degradation on a busy short-haul cable that downstream traffic engineering can route around within minutes.
The 16 high-severity events
Every alert classified as critical during the window:
| Date (UTC) | Cable | Segment | RTT baseline → current (ms) | Resolution time |
|---|---|---|---|---|
| 2026-04-07 02:32 | COBRAcable | Eemshaven, NL → Endrup, DK | 34.0 → 44.7 | 178 min |
| 2026-05-02 11:01 | Tannat | Las Toninas, AR → Santos, BR | 341.1 → 26.9 | 120 min |
| 2026-05-02 03:01 | Tannat | Las Toninas, AR → Santos, BR | 290.0 → 27.2 | 210 min |
| 2026-04-30 23:01 | Tannat | Las Toninas, AR → Santos, BR | 249.2 → 232.4 | 120 min |
| 2026-04-29 09:01 | Tannat | Las Toninas, AR → Santos, BR | 189.0 → 27.1 | 89 min |
| 2026-04-28 15:00 | East-West Cable System | Mersing, MY → Penarik, ID | 34.7 → 556.3 | monitoring (open) |
| 2026-04-28 05:01 | Tannat | Las Toninas, AR → Santos, BR | 156.1 → 125.8 | 209 min |
| 2026-04-17 11:02 | Tannat | Las Toninas, AR → Santos, BR | 81.7 → 27.0 | 119 min |
| 2026-04-17 05:01 | PGASCOM | Sakra Island, SG → Kuala Tungkal, ID | 55.0 → 87.8 | 90 min |
| 2026-04-16 09:01 | NCSCS | Kribi, CM → Lagos, NG | 49.9 → 20.6 | 298 min |
| 2026-04-15 19:01 | PGASCOM | Sakra Island, SG → Kuala Tungkal, ID | 43.3 → 84.9 | 89 min |
| 2026-04-15 05:01 | BaSICS | Batam, ID → Kuching, MY | 8.5 → 5.3 | 576 min |
| 2026-04-13 09:01 | B2JS | Tanah Merah, SG → Jakarta, ID | 42.7 → 83.0 | 90 min |
| 2026-04-08 02:32 | Southern Caribbean Fiber | Chaguaramas, TT → San Juan, US | 59.1 → 44.8 | 419 min |
| 2026-04-05 22:30 | Blue | Aqaba, JO → Marseille, FR | 79.0 → 65.9 | 60 min |
| 2026-04-04 06:00 | GO-1 Mediterranean | St. Paul’s Bay, MT → Mazara del Vallo, IT | 44.6 → 41.7 | 197 min |
| 2026-04-04 05:00 | SJC | Tuas, SG → Chikura, JP | 106.9 → 122.7 | 257 min |
| 2026-03-11 00:33 | B2JS | Tanah Merah, SG → Jakarta, ID | 43.0 → 95.8 | 1168 min |
One row demands a closer look. The April 28 East-West Cable System alert — Mersing, Malaysia → Penarik, Indonesia — recorded a current measurement of 556 ms against a 35 ms baseline. That is a 16× deviation, and unlike everything else in the table, the alert never moved to resolved. It remained in monitoring state — meaning subsequent measurements never returned cleanly to baseline within our auto-resolve window. Five days later, at the time of writing, it is still open. Whether this is a measurement-side problem (a probe drift, a route change at the IP layer that puts our prober behind a slower path) or a cable-layer problem we have not yet disambiguated. Either way, it is the longest unresolved high-severity event of the window.
Where the events cluster
The 49 affected cables are not distributed uniformly. Five geographic clusters account for the majority of events:
| Cluster | Cables involved | Approximate event count | Notable patterns |
|---|---|---|---|
| Southeast Asia data triangle (SG/ID/MY) | B2JS, BDM, SEAX-1, BaSICS, PGASCOM, Apricot, BRCS, DMCS, RISING 8 | ~28 | Highest volume of any region. Short-haul cables connecting Singapore data hubs to Indonesian and Malaysian terminals. Most events resolve quickly; route diversity through alternate cables masks them at the application layer. |
| Mediterranean basin | GO-1 Med, Italy–Malta, KELTRA-2, Ionian, Italy–Albania, Adria-1, Silphium | ~12 | Small enclosed basin, high cable density, limited redundancy on individual island/peninsula links. The Greek and Italian terminals show repeated mild degradations. |
| South Atlantic (AR–BR–UY) | Tannat, Firmina | ~6 | Tannat dominates with five high-severity events. Las Toninas → Santos is one of the busiest financial-traffic corridors in the southern hemisphere; recurring degradation here matters disproportionately. |
| Africa west coast | NCSCS, Maroc Telecom West Africa | 2 | NCSCS Kribi–Lagos saw a high-severity event April 16; Maroc Telecom Libreville–Casablanca a warning April 25. Limited measurement coverage but each event affects entire countries. |
| Trans-Atlantic and Trans-Pacific long-haul | AAG, AC-1, FA-1, JUPITER, TPE, TPU, ASE/Cahaya Malaysia | ~10 | Long-haul cables saw mostly warning-tier events, suggesting degradations rather than dramatic spikes. Multiple alerts within the same hour on AAG and B2JS on April 5 and April 7 hint at coordinated upstream events — possibly maintenance or routing changes propagating. |
Resolution time tells its own story
Across all 82 events, the median time from detected to resolved was 204 minutes — a little under three and a half hours. The fastest 25% resolved in 89 minutes or less. The slowest 25% took longer than 419 minutes. A handful sat in monitoring for far longer:
- TPU (Claveria, PH → Eureka, US): 8,607 minutes / 5.97 days — a transpacific warning that never dropped back to baseline cleanly.
- BRCS (Rengit, MY → Tanjung Pinggir, ID): 6,024 minutes / 4.18 days — an Indonesia–Malaysia short-haul stuck in degraded state.
- Mariana–Guam Cable (Saipan, MP → Tanguisson Point, GU): 2,410 minutes / 1.67 days.
- East-West (Mersing, MY → Penarik, ID): still open at the time of writing, 5+ days.
The 89-minute floor on most resolutions is, we suspect, a hard signal. Below that, our system requires three consecutive clean measurement cycles before declaring resolution, and our measurement cadence runs roughly every 25–30 minutes per route. So 89 minutes is essentially the system saying: by the time we could possibly mark this resolved, it was already gone. These events likely correspond to upstream routing decisions — BGP shifts, MPLS path changes, traffic engineering — that flap and recover within a single human attention span.
What the data does not show
Cable cuts are absent from this dataset. The Red Sea events of 2024, the Baltic incidents of 2023, the Hunga Tonga submarine landslide of January 2022 — none of these would appear in a 60-day rolling window from March–April 2026. What you see here is the everyday weather of submarine cables: small degradations, transient spikes, and a few persistent open questions. None of these 82 events made international news. None of them caused widespread internet outages. Every one was a probe-level signal that something, somewhere, was less than perfect for a few minutes to a few hours.
Reading the events in aggregate, two operational observations stand out. First, recurring instability is far more common than single dramatic spikes. Of the 49 affected cables, 7 produced three or more events — a small minority responsible for a disproportionate share of the alerts. Second, almost everything resolves on its own. Whether that is because the underlying issue physically corrects (sea-state shifts away from a vulnerable route, a router rebalances) or because traffic engineering routes around the degradation before users notice, our data cannot say. The event simply ends.
Methodology and limitations
Our measurement coverage favors cables landing in or routing through cities where we have probe density: Singapore, Marseille, Frankfurt, Tokyo, Mumbai, São Paulo, Buenos Aires, Lagos, and a handful of others. Cables landing exclusively in countries where we have no probe coverage will not generate alerts in this dataset, regardless of their actual condition. Caribbean cables, smaller Pacific Island links, and parts of the Arctic backbone are under-represented for this reason. We are expanding probe coverage continuously; the next wave includes additional anchors in West Africa and the South Pacific.
Severity classification combines absolute RTT increase, persistence across measurement cycles, packet-loss correlation, and route-fingerprint changes detected via traceroute. A single anomalous ping in isolation does not trigger an alert. The thresholds are tuned to keep the warning rate at roughly one event per cable per quarter under normal conditions, with critical reserved for events that would be visible to a network operator looking at their own monitoring.
The full dataset behind this report — every alert, every measurement timestamp, every probe ID — is queryable through the GeoCables admin panel. We publish a refreshed health report every 30 days; the next one will cover the May–June 2026 window.
