The Squid’s InkSquidGrid’s weekly read of the internet — fresh ink every Monday

Week 37, 2026 · Aug 31 – Sep 6

Covers Monday Aug 31 00:00 to Monday Sep 7 00:00 UTC.

Correction. Restated on Sep 16 for the standard Monday-to-Monday week, so each issue covers one full week and compares with the issue before it.

How to read this: we measure millions of internet routes around the world. Every speed number below is how many milliseconds (thousandths of a second) slower a route ran than its own normal speed, so higher is worse. Each section says whether things got better or worse than last week.

1,837,841internet routes we measure

736,783,472measurements taken this week down 1% from last week

1,505outages and problems companies reported

533of them rated major or critical

Bharti Airtel Limited’s rough weekthe week’s story

Better than last week. Bharti Airtel Limited is one of the large networks that carry internet traffic between countries. Routes through it ran 34.0 milliseconds slower than normal on average this week (last week: 35.6). 1,957 routes through it were badly slowed (more than 25 milliseconds over normal) at some point. The chart shows how many, hour by hour:

Signed-in customers see the affected routes, the carrier on each side of every changed handoff, and the regional split on the Global fleet view.

The networks that slowed down the most

Large carrier networks, ranked by how much slower than normal their routes ran this week. Big networks get no allowance for their size.

Network	slower than normal	routes badly slowed	last week	compared with last week
SG.GS (Singapore network operator)	46.6 ms	836	44.0 ms	Worse than last week
Bharti Airtel Limited	34.0 ms	1,957	35.6 ms	Better than last week
FLAG Telecom	28.3 ms	268	28.1 ms	About the same as last week
Gateway Communications (Belgium)	27.3 ms	2,743	—	New to this list this week
TransTelecom	27.2 ms	538	27.1 ms	About the same as last week
Telxius	25.9 ms	1,893	—	New to this list this week

“Routes badly slowed” counts routes that ran more than 25 milliseconds over normal at some point this week.

1,505 outages and problems reported by companies

These are problems online services posted on their own status pages. This week’s count can’t be compared with last week’s: 680 companies reported problems this week against 101 last week, because we began following many more status pages.

SaaS990

Payments & fintech159

Education & research73

AI68

Crypto44

Cloud & hosting42

Consumer33

Security & identity33

Online business software (SaaS) had the most problems again, as it usually does.

What tended to break, in providers’ own words: connectivity ×174 · slowness ×156 · auth & login ×103 · elevated errors ×78

Longest problems to fix: Imperva — [INC-1106] Elevated Latency for a Subset of Origin Fetch Traffic in As (481 hours, about 20 days); Loftware Cloud — Health Checks issue on Cloud Automation (468 hours, about 20 days); Kilo — Usage page is not showing any data for users. (459 hours, about 19 days).

Ground eventsthe physical world, and whether the internet flinched

Mar 1 – Apr 30 · six-month check-in The Gulf cloud regions, half a year after the strikes
The drone strikes that damaged AWS facilities in the UAE and Bahrain are March's news, not this week's — but they remain the region's defining infrastructure fact, so we check in. AWS ran 'Increased Error Rates' events for both regions for eight straight weeks and closed them on April 30; its health board has been clear since. What the board can't show is the road back.
Our fleet’s view: 2,032 paths from the UAE ran +9.6 ms over their own norms this week, Bahrain +15.1 ms — mildly elevated, no incident on the board. Whether that residue is reduced regional capacity or ordinary variance, path data alone can't say. What it can say: the recovery has held.
Cloudflare Q2 disruption review · TechPolicy.press on the legal fallout

This week · quiet The planet cooperated
No earthquakes reached M7.0 (USGS), and no storm reached GDACS red or orange alert. Tropical Depression 23 in the NW Pacific gets one sentence only because its track brushes the Luzon Strait cable corridor — nothing measured, nothing expected yet.
USGS feed · GDACS

Ongoing, week 12 of tracking: Iran's government-directed blackout: selective whitelisting continues; where our 80 Iranian paths are reachable at all, they ran +12.8 ms over norms this week.

We cover earthquakes of magnitude 7.0 or more and storms at the top two disaster alert levels. “Our fleet’s view” is what our own measurements saw, not a damage report.

Subsea watchwhere the internet is wet

Sep 8 · new Vocus fault on the Australia–Singapore corridor
A fault on the Australia–Singapore Cable was reported Monday and remains under repair. The interesting part is what didn't happen: no corridor blackout.
Our fleet’s view: 8,868 paths from Australia averaged +12.8 ms over their norms in the 48 hours after the fault, and 12,053 from Singapore +14.3 ms — the signature of traffic squeezing onto surviving systems, not going dark. Singapore was also this week's worst measuring region for Microsoft Azure paths (+31 ms): same corridor, second lens.
SubTel Forum

Since Aug 27 · ongoing Vietnam is still down four international cables
Two weeks in, the concurrent faults keep international capacity constrained — on paper, one of the worst connectivity stories running.
Our fleet’s view: The 646 paths we measure from Vietnam averaged just +6.3 ms over norms. Domestic rerouting is carrying it far better than the headlines suggest.

This week · wire Repairs and consequences
Rostelecom restored the Kamchatka–Sakhalin link (Aug 18); India moved to fast-track repair-ship clearances (Aug 26) — the quiet policy change that shortens every future fault in the region; and the trial over last winter's Baltic cable damage opened in Finland on Sep 8.
SubTel Forum

China gateways

China Telecom 48.7% / norm 51.2%

China Mobile 91.4% / norm 91%

China Unicom 75.5% / norm 78.4%

Reachability of measured routers inside each network, seen from outside the gateway, against each network’s own trailing norm. The lines wiggle; that is the point.

For the feedpaste-ready

The Squid's Ink — Week 37. The internet, measured, not guessed. The cable story: Vocus lost the Australia–Singapore run Monday. Our fleet's answer: +13–14 ms from either end — rerouted, not dark. Vietnam, four cables down since late August, is holding at +6 ms. The backbone story: Bharti Airtel had the week's roughest ride — 1,918 paths over their norms. League table in the report. The long shadow: six months after the Gulf data-center strikes, AWS's board is green but our UAE paths still sit +9.6 ms over baseline. 1,247 provider incidents archived this week. Every number is a departure from that path's own norm, measured by SquidGrid's independent fleet. Full report: https://squidgrid.net/reports

Every figure derives from SquidGrid’s own measurement fleet and its archive of provider-published incidents; provider self-reports are labeled as such. Provider Assessment Reports carry the per-provider detail. The internet was mostly fine this week. We measured anyway.