Meta, AWS EC2 Blips Recovered
No widespread internet outage today. As of about 09:15 UTC on September 22, core cloud, CDN and DNS infrastructure is operating normally. Two notable but short incidents overlapped on Sunday evening US time, and both have recovered.
Facebook and Instagram stopped loading for many US users from around 8:55 PM ET on Sunday, September 21 (00:55 UTC Monday). Downdetector logged more than 17,000 Facebook reports and over 5,000 for Instagram at the peak, and some users outside the US were affected too. Reports fell sharply by about 10 PM ET and service was normal by Monday morning. Meta has not given a cause, and its business status page at https://metastatus.com/ shows nothing open. This is the third notable Meta disruption of 2026, after larger ones in June and July.
AWS had a control-plane problem in its busiest region at the same time. Between 22:49 and 00:06 UTC (3:49 to 5:06 PM PDT on September 21), new EC2 instance launches in us-east-1 (N. Virginia) failed or returned errors through RunInstances, CreateFleet and related APIs. Services that launch instances on customers' behalf, including ECS, Fargate and EMR, were also affected. Already-running instances were not impacted. AWS attributes it to a recent deployment to a subsystem that processes instance launches, and marks the event resolved. Details at https://health.aws.amazon.com/health/status
Cloudflare closed two brief, minor edge incidents this morning: network connectivity in London (LHR) from 02:25 to 03:36 UTC and performance in San Jose (SJC) from 07:20 to 08:20 UTC. On Sunday afternoon, R2 PUT requests saw elevated failures in the WNAM and ENAM regions from 15:25 to 18:00 UTC, and Containers briefly could not start in Asia-Pacific. The low-impact WARP geolocation issue open since August 27 is still under repair. See https://www.cloudflarestatus.com/
One item is still open. Azure DevOps reports Test Plans degraded in Europe, with Microsoft "investigating a service disruption" as of 09:00 UTC. Core services, Repos, Pipelines and Boards are healthy in every region, and the main Azure status page shows no active events. Track it at https://status.dev.azure.com/
The AI platforms had a busy night but are clean now. Anthropic reported elevated error rates across several Claude models from 00:50 to 02:10 UTC, resolved after the root cause was found (status.claude.com). Cursor resolved a degradation affecting Grok model requests at 01:42 UTC, its second Grok-related incident in two days. OpenAI, Groq, Replicate, ElevenLabs, Cohere and Stability AI report no incidents.
Smaller items: Akamai has an advisory for elevated latency in Singapore, Hong Kong and Japan during upstream provider maintenance that runs through October 3; core delivery is unaffected. Twilio lists SMS delivery delays to Airtel in India, Du in the UAE and Telemat in Brazil, plus continuing voice call failures to Hong Kong, all carrier-level rather than platform-wide. Zoom resolved an Apple sign-in problem for a subset of users on Sunday. Earlier in the 72-hour window, GitHub had delays creating merge commits for about an hour on Saturday (22:13 to 23:22 UTC), and Vercel cleared three deployment-related incidents on Thursday.
All clear at Google Cloud, Azure, GitHub, Discord, Atlassian, Slack, Vercel, Heroku, DigitalOcean and Stripe. No BGP or DNS-level events of note, and no new submarine cable damage reported.
Bottom line: two short, contained incidents on Sunday evening, one minor regional item open at Azure DevOps, nothing internet-wide. The internet is up.