An AWS outage degrades S3 and DynamoDB and drags fifteen services down with it
Since 14 November 2026, Amazon S3 and DynamoDB have been degraded, and AWS says it is restoring availability through software mitigations under infrastructure constraints. By the morning of 15 November, fifteen of the sixteen services tracked by third-party observers were affected: isolate your S3 and DynamoDB dependencies and revisit your regional failover runbook.
14 November 2026. The AWS Health Dashboard flips to an incident on Amazon S3 and Amazon DynamoDB, its two foundational services. The next morning, at 9:00 UTC, third-party tracker awsdown.com counts 15 of 16 monitored services as degraded, with only EBS still operational. The wording of AWS’s statement is worth reading closely: the vendor says it is “restoring data access and service availability through software mitigations, including deploying updates to enable S3 to operate within the current infrastructure constraints”. Why it matters: when S3 and DynamoDB stumble, everything built on top of them — Lambda, CloudWatch, IAM, Route 53, RDS — stumbles too. This is the architecture lesson of the week, not just another entry in the outage history books.
Two foundational services, one star-shaped cascade
The blast radius reads like a dependency list. Degraded, according to the awsdown.com snapshot taken on 15 November at 9:01 UTC: S3, DynamoDB, EC2, Lambda, EKS, RDS, ElastiCache, CloudWatch, SNS, SQS, API Gateway, CloudFront, Route 53, VPC, and IAM. EBS, the block storage service, is the only one of the sixteen tracked services still operational.
None of this is surprising. S3 is not just an object store: it is where Lambda loads function code, where CloudWatch writes its exports, where CloudFront reads its origins. DynamoDB hosts the control-plane tables of many managed services — queues, locks, service metadata. And IAM plus Route 53 form the control plane through which nearly every API call is authenticated and nearly every name is resolved. When the foundation shakes, every floor shakes with it.
That is exactly what AWS’s sentence describes. “Software mitigations” to let S3 “operate within the current infrastructure constraints” describe a situation where the problem is not a simple application bug: the software must adapt to degraded infrastructure underneath it. In other words, a physical resource — network, power, cooling, or storage hardware — is constrained, and the software layers are being reconfigured to keep running in spite of it. AWS has not, at the time of writing, published a public root cause or a single identified region.
What “degraded” means for an SRE
“Degraded” is not “down”. A degraded service still accepts requests, but with higher latency, intermittent errors, partial failures on some operations or in some regions. For an SRE, that is often more expensive than a hard outage: a hard outage trips your failover mechanisms, whereas a diffuse degradation poisons your metrics, fills your retry queues, and inflates your costs without ever crossing a single clean alerting threshold.
The most insidious vector here is the indirect dependency. Your application may never talk to DynamoDB directly. But if it runs behind API Gateway, Lambda, or SQS, each of those links can slow down or fail because DynamoDB or S3 is slow upstream. Your dashboard will show errors on your service without ever pointing at the foundation causing them. So the first useful reaction is not to redeploy your application, but to correlate your errors with the status of upstream dependencies — and to read AWS’s eventual post-event summaries, not the aggregators, to separate signal from noise.
The architecture lesson: treat S3 and DynamoDB as Tier 0
The real question this incident poses is not “when will AWS fix it”, but “why does your continuity plan assume S3 and DynamoDB never fail”. The most common answer is the SLA: S3 and DynamoDB carry some of the highest availability commitments in the catalog, and a decade of operations has installed the belief that they are “too big to fail”. A 99.99% SLA still leaves 52 minutes of downtime a year, and it does not cover the partial degradations that trigger no credit at all.
The practical consequence is that many architectures concentrate their resilience in the wrong place. They multi-AZ the application, but keep an implicit regional dependency on S3 or DynamoDB: a single table in a region, a single bucket for shared state, leadership locks stored in DynamoDB with no fallback path. When one region’s foundation degrades, multi-AZ saves nothing — it only spreads the failure across three zones that are equally affected. The resilience that matters against a foundation incident is cross-regional, not intra-regional.
This is the moment to take inventory. For every critical workload, list your S3 and DynamoDB dependencies and ask: what happens if this service degrades for four hours in my primary region? If the answer is “we wait”, your continuity plan has a hole. Acceptable answers are cross-region replication (for DynamoDB), bucket replication, or an explicitly designed degraded mode that runs without shared state for the duration of the incident.
The “foundation is down” runbook nobody has
Few teams have a runbook for “S3 is degraded”. It is nonetheless the highest-value exercise of the week. The runbook does not need to be long, but it must settle three questions in advance, because none of them are settled well in the heat of the moment.
First, detection. How does your monitoring distinguish a S3 or DynamoDB degradation from a bug in your own application? If you run on AWS, correlate your alerts with the AWS Health Dashboard and your account’s Personal Health Dashboard feed — that is the source that knows whether your account is affected, not just the cloud at large.
Second, failover. How long does it take to point your application at another region, and is that number written down anywhere? For DynamoDB, cross-region replication fails over in minutes if it is already configured; if it is not, failover is measured in hours, the time to reindex a table from a backup.
Third, throttling. Which queues, retries, and exponential backoff loops will you cut to stop a degradation from becoming an avalanche of requests against an already crippled foundation? Cutting aggressive retries is often what turns a long incident into a short one, for you as much as for your neighbours on the same foundation.
This is not the first time S3 has taken a chunk of the internet down with it. The February 2017 us-east-1 S3 outage knocked out a wide swath of dependent services for roughly four hours and was later estimated at around $150 million in losses. The May 2026 us-east-1 thermal event stretched to roughly 28 hours. The recurrence of the same pattern, a foundational service dragging its dependents down, is itself the strongest argument for treating S3 and DynamoDB as Tier 0 rather than as invisible plumbing.
The verdict
If you run production workloads in an affected AWS region, start by correlating your errors with the status of S3 and DynamoDB before you redeploy anything: the odds that your bug is actually a foundation degradation are high. Check your account’s Personal Health Dashboard, then apply your throttling playbook — cut retries, degrade non-critical functions cleanly.
If you discover that your architecture implicitly assumes S3 or DynamoDB never fails, do not simply wait out the incident. Turn on DynamoDB cross-region replication and replication for your critical buckets, write the “foundation is down” runbook, and rehearse it once. The cost of an unreplicated dependency on the foundation comes due exactly on the day the foundation gives way — which is today.