AWS Outage October 2025: What Caused It and How to Future-Proof Your Business

AWS logo over a cloud with warning icons symbolizing the October 2025 outage.

AWS still holds the largest share of global cloud infrastructure spending: 26% in Q3 2026, ahead of Microsoft at 19% and Google at 16%, according to Synergy Research Group. At that scale, a fault in one Virginia facility reaches UK banks and government services within minutes. On 20 October 2025, customers of Lloyds, Halifax and Bank of Scotland could not log in, and HMRC’s online services went down.

Key Takeaways From Recent AWS Outages

  • AWS has had two headline outages in seven months, both in US-EAST-1, from unrelated causes.
  • October 2025: a software fault took out the whole region. Only multi-region designs with an independent control plane stayed online.
  • May 2026: overheating knocked out one Availability Zone. Multi-AZ designs survived it.
  • A resilience plan needs a tested answer for both kinds of failure.

AWS Outage Timeline

DateRegionScopeRoot CauseDuration
7–9 May 2026US-EAST-1Single AZ (use1-az4)Thermal event causing loss of power to EC2 and EBS24+ hours
19–20 Oct 2025US-EAST-1Whole regionLatent race condition in DynamoDB DNS automation~15 hours

Last updated: 14 November 2026.

October 2025: The DynamoDB DNS Failure

At 23:48 PDT on 19 October 2025 (07:48 BST on 20 October), DynamoDB in US-EAST-1 began rejecting requests. Snapchat, Alexa, Coinbase, Duolingo and Fortnite failed within minutes, followed by UK banking apps as the working week began.

How the Failure Unfolded

Two automated DNS Enactors applied conflicting plans to DynamoDB’s regional endpoint, and a cleanup job then deleted the record still in use. With no IP addresses behind dynamodb.us-east-1.amazonaws.com, every client lost its connection, AWS’s own internal services included. Engineers isolated the DNS state as the source within 50 minutes and restored the record in under three hours.

Recovery then stalled further down the stack: EC2 could not launch new instances, and Network Load Balancer health checks failed, so services built on them stayed degraded until 14:20 PDT on 20 October.

AWS logo over a cloud with warning icons symbolizing the October 2025 outage.

How a DNS Failure Triggered a Global Cloud Disruption

An empty DNS record cut off every system that called DynamoDB in the region. Many core AWS services, from Identity and Access Management to CloudFront, rely on US-EAST-1 for coordination.

When that single point of resolution went dark, it triggered a chain reaction of cloud service disruption worldwide.

Even companies running workloads in other AWS regions weren’t fully protected. Many “multi-region” setups still depended on US-EAST-1 for authentication or routing. As a result, the AWS downtime rippled far beyond one data centre, revealing just how interconnected the cloud truly is.

Inside the Aftermath: Recovery and Realisations

By late morning, most major platforms were back online, but the recovery process was gradual. AWS engineers spent hours draining backlogged requests, recalibrating load balancers, and verifying data integrity. 

AWS’s post-event summary traced the trigger to a latent race condition in the DynamoDB DNS management system, and AWS disabled the DNS Planner and Enactor automation worldwide while it added safeguards. Uptime Institute’s analysis of the incident found that applications architected across regions kept running, while single-zone and multi-zone deployments inside US-EAST-1 went down with it.

May 2026: The US-EAST-1 Thermal Event

On 7 May 2026, a cooling failure in one facility inside Availability Zone use1-az4 forced servers to shut down. EC2 instances and EBS volumes in that zone stayed impaired until 9 May, and Coinbase and FanDuel both reported trading disruptions.

This time, zone design decided the outcome. Teams with capacity in other zones moved workloads out within hours, and several platform providers failed over replicated namespaces before customers noticed. Workloads pinned to one AZ had nowhere to go.

One detail caught even prepared teams. AWS maps zone names such as us-east-1c to different physical zones in each account, while AZ IDs such as use1-az4 stay consistent. Runbooks written with names alone sent engineers hunting for the wrong zone mid-incident.

The Domino Effect: How Businesses Felt the Impact

When the AWS outage hit, it sent shockwaves through the digital economy. Millions of users suddenly couldn’t stream, shop, or play, revealing the massive impact the outage had on apps that drive everyday life. 

UK Banking and Government Services Went Down Too

Reports of UK outages spiked for Lloyds, Halifax, Bank of Scotland, HMRC, Gov.uk, and the major mobile networks, and Lloyds Banking Group confirmed the link to AWS. A fault in a US region interrupted UK retail banking during the Monday morning rush, with nothing wrong in any UK data centre.

The True Cost of Cloud Downtime for Global Businesses

AWS compensates SLA breaches with service credits against the affected bill. Lost orders, failed payments and support overtime stay with the customer. Regulated UK firms carry an extra cost: showing the FCA and PRA that important business services stayed within their impact tolerances.

Customers also aim their anger at the brand on their screen. Few of them know which cloud provider runs it, and fewer care.

Why Relying on a Single Cloud Provider Is a Growing Risk

The outage underscored a painful truth: too many organisations depend entirely on one cloud provider. AWS’s dominance has created a convenience trap, powerful tools, but a centralised vulnerability.

When the AWS US-EAST-1 region went offline, companies without multi-cloud redundancy or cross-region backups were instantly paralysed. This event showed that vendor lock-in is more than a business risk; it’s a potential operational disaster.

11 Lessons Learned: How to Build Cloud Resilience Before the Next Outage

  1. Match the failure domain to the workload: Multi-AZ absorbs facility failures like the May 2026 thermal event, while only multi-region absorbs control-plane failures like October 2025. Run active-active or active-passive in at least two regions; keep DNS, secrets, CI/CD runners, and observability out of a single region. Prioritise services with native cross-region options (DynamoDB Global Tables, Aurora Global Database, S3 CRR, ECR replication, MSK multi-region patterns). Target RTO < 30–60 min and RPO ≈ 0–5 min for tier-1 systems.
  2. Harden DNS and endpoint resolution: Use dual DNS (e.g., Route 53 + a secondary like Cloudflare) with health checks and latency/geo routing. Keep low TTLs (but not so low you hammer resolvers). Prefer regional service endpoints and failover records; avoid hidden dependencies on us-east-1 for control traffic.
  3. Decouple the control plane from the data plane: Host CI/CD, IaC state backends, artefact registries, and runbooks so they stay reachable during AWS downtime. For identity, prefer external IdPs with regional STS endpoints; don’t let auth for your whole estate hinge on a single region.
  4. Replicate data intentionally: Choose the replication mode per workload: async (eventual), sync (quorum), or log shipping. Define RPO per domain (orders ≈ 0, analytics can tolerate minutes). Validate cross-region schema drift and failover playbooks for stateful stores (caches, queues, search clusters).
  5. Bake in application-level resilience: Standardise timeouts, exponential backoff with jitter, and circuit breakers. Make calls idempotent. Add bulkheads between features so one failing dependency doesn’t cascade. Implement graceful degradation: read-only mode, queue writes, disable non-critical features, serve cached content.
  6. Engineer for dependency independence: Inventory external APIs, SaaS, and “global” AWS services. Provide stub/fallback paths (e.g., alternate payment provider, secondary email/OTP). Prefer regionalised alternatives to global endpoints where possible.
  7. Operationalise failover (don’t just document it): Run game days and chaos drills quarterly: break DNS, kill a region, throttle a dependency. Measure SLOs, alert on error budgets, and practice push-button failover. Keep a provider-independent status page and comms playbook.
  8. Control costs without sacrificing resilience: Use warm standbys or pilot-light architectures for non-critical tiers. Right-size replica capacity, scale up on failover. Automate teardown after recovery to avoid “resilience tax” lingering.
  9. Reduce vendor lock-in where it matters: Abstract at the CI/CD and platform layer (containers, Terraform modules, OPA policies). Standardise on 12-factor app patterns and portable observability. If multi-cloud isn’t feasible, build cloud-portable so you can move the top 20% of revenue-critical services first.
  10. Make resilience visible to the business: Tie risks and budgets to business capabilities (checkout, auth, streaming). Report RTO/RPO, failover drill results, and outage blast radius in exec-friendly terms. Resilience is a product feature; market it.
  11. Write AZ IDs into runbooks: Record use1-az4-style IDs alongside zone names, so incident responders can match AWS Health Dashboard notices to their own infrastructure in seconds.

Immediate Cloud Resilience Actions to Take Today

  • Two regions defined per tier-1 system; DNS failover ready: Run your most critical systems across multiple regions with automated DNS rerouting in case one fails.
  • Data replication RPO/RTO documented and tested: Know exactly how much data you can afford to lose (RPO) and how fast you can recover (RTO), and validate it through real tests.
  • Timeouts/backoff/circuit breakers enforced at SDK/middleware level: Build defensive logic so your apps degrade gracefully instead of collapsing under dependency failures.
  • External IdP + regional STS endpoints configured: Prevent login failures by using external identity providers and regional authentication endpoints.
  • Quarterly chaos drills scheduled; comms runbook and independent status page live: Practice outage scenarios and maintain a communication plan that stays online even when AWS isn’t.

AI dependencies follow the same rules, as the Claude outage tracker shows for Anthropic incidents.

Why These Outages Keep Happening

Hyperscale clouds fail less often than most private data centres, and they fail bigger. Shared systems for DNS, identity and orchestration let one fault, whether a stale DNS plan or an overheated data hall, reach thousands of customers in minutes.

Why US-EAST-1 Fails Louder Than Other Regions

At the heart of the issue is AWS region dependency. Despite operating dozens of data regions worldwide, many of Amazon’s critical services still lean on US-EAST-1, the first and most interlinked of them all. It’s where authentication, DNS, and routing often converge, a silent backbone that most companies don’t even realise they’re relying on.

Apps in Europe or Asia might not host their data there, yet they can still freeze if authentication or API calls are routed through Virginia.

When the Control Plane Fails, Everything Fails

Think of the control plane as the brain of AWS; the system that tells every component when to start, stop, scale, or connect. When it jams, the body keeps its organs but loses coordination. That is what happened in October 2025: running instances stayed up while new EC2 launches and load balancer health checks failed.

Deployment pipelines, monitoring tools and auto-scaling triggers stall alongside them, because they depend on the same centralised logic.

DNS: The Small Service That Can Break the World

October’s empty DNS record shows how much can rest on a single entry. When DNS fails, it blinds entire systems. Services can’t find databases, APIs can’t locate authentication endpoints, and recovery scripts can’t even start because the paths to them no longer exist.

Redundancy Isn’t the Same as Resilience

Multiple instances in one region protect against a failed server and, when spread across zones, a failed facility. They offer no protection when the region’s control plane breaks.

Full redundancy means spreading workloads across fault domains and decoupling identity and DNS from any single region. It costs more, so it rarely wins budget until an outage forces the conversation.

A Pattern That Keeps Repeating

AWS has weathered outages like this in 2017, 2021, and 2023, each one following a familiar pattern: regional interdependence, internal service overlap, and DNS propagation delays. 

Engineers fix the immediate cause, but the underlying design remains; massive complexity is centralised around a few critical points.

Other providers show the same pattern. A configuration change took down Microsoft Azure Front Door on 29 October 2025, and a Google Cloud failure in London’s europe-west2 region disrupted Revolut on 20 October 2026, as logged in the GCP outage tracker.

The Real Question

Two major US-EAST-1 failures in seven months, from unrelated causes, settle whether AWS will fail again. The question for a CTO is which failure domain each tier-1 service can survive today, and when that answer was last tested.

Business Lessons from the AWS Outage

The world’s largest brands run on the same cloud foundations as small startups, and when those foundations shake, size offers no protection.

Downtime Is No Longer Just an IT Problem

When services go offline, the damage reaches far beyond the IT department. Sales pipelines freeze, customer support is overwhelmed, and marketing campaigns lose momentum. The organisations that recover fastest have rehearsed recovery before they need it.

Even with Amazon’s own commitment to a 99.99% monthly uptime for EC2 instances deployed across multiple Availability Zones or regions, that remaining 0.01% still translates into potential downtime, enough to disrupt entire customer journeys when architecture isn’t built for recovery.

Multi-Cloud and Hybrid-Cloud Are Now Essential

Multi-region AWS covers every failure mode seen in the two recent outages. A second provider earns its cost where a provider-wide fault would be unacceptable, or where a regulator expects a credible exit plan, as the PRA does for material outsourcing.

A strong multi-cloud or hybrid-cloud strategy ensures that critical functions keep running even when one vendor fails. For example, identity management can live on Azure, storage on AWS, and analytics on Google Cloud.

Business Continuity Depends on Testing

Having a disaster recovery plan written in a document isn’t enough. Business continuity cloud strategies must be tested regularly and updated as systems evolve. Run quarterly failover drills, practice DNS rerouting, and simulate communication plans for customer updates.

Uptime Institute’s July 2026 availability analysis, built from provider status histories, quantifies the gap: in the worst case, a single-zone application would have lost 21.5 hours in 2025 (double the 2024 figure) while the worst-hit multi-region application lost about six hours.

Teams that test recovery workflows already know how long it takes to switch regions or restore backups, and they identify gaps before real customers notice them.

Cyber incidents across the UK, from JLR to Heathrow, reinforce the same principle: testing and proactive DevSecOps can prevent massive financial losses during disruptions, as outlined in this analysis of 2025 cyberattacks.

Shift Focus from Cost Optimisation to Risk Optimisation

Many businesses have designed their cloud architecture around cost efficiency. The AWS outage shifts that mindset.

Next to “How much does it cost to run?”, the board needs a second figure: “How much does it cost to fail?” Cutting redundancy saves money until the first long outage.

Forward-thinking companies now prioritise risk optimisation, balancing cost with performance, redundancy, and resilience to protect their revenue and reputation.

Cloud Resilience as a Competitive Edge

Resilience used to be an engineering goal; now it’s a market differentiator. Clients, partners, and investors all ask one question: “Can you stay online when others can’t?” 

The businesses that can confidently answer yes, supported by a tested cloud resilience strategy, will have the upper hand in trust and reliability.

If your business runs in the cloud, plan for the day it doesn’t. Those who treat resilience as optional will keep learning the same lesson the hard way.

How to Future-Proof Your Business

Cloud uptime is a shared responsibility: AWS runs the facilities and services, and each customer decides how much failure its architecture can absorb. The following steps will help any organisation build a cloud management and resilience strategy that stands firm when providers stumble.

  1. Build for Multi-Region and Multi-Cloud Continuity

Resilient businesses design for flexibility. That means running workloads across several AWS regions or even multiple cloud providers to mitigate cloud provider risk. A failure in one region should never stop critical operations in another.

Deployflow helps companies do exactly that through Cloud Consulting and Cloud Migration services, mapping out architectures that stay compliant, scalable, and ready to shift under pressure. These strategies turn single-region dependence into multi-region confidence.

  1. Automate Failover and Continuous Monitoring

Manual recovery belongs to another era. Today, every second counts. Automation should detect failures, reroute traffic, and restart workloads without waiting for human input. Continuous monitoring closes the loop, giving teams full visibility of performance, costs, and security.

With cloud management services, you’ll enable automatic failover and predictive monitoring that prevent outages from becoming downtime.

  1. Strengthen the Edges: DNS, CDN, and Access Layers

Even the most redundant computer infrastructure can crumble if everything depends on one DNS or CDN provider. Diversifying these services keeps users connected when one network falters. The same goes for access controls and encryption; every entry point needs its backup plan.

Cloud Security focuses exactly on that layer of protection, combining multi-provider DNS setup, zero-trust principles, and automated compliance monitoring to create a defence that’s invisible until it needs to act.

  1. Test Backup and Recovery Plans Regularly

A recovery plan is useless if it only lives in a document. Resilience comes from regular testing: restoring data, failing over regions and confirming the backup plan works under load.

Deployflow integrates these routines into everyday workflows through its consulting and management frameworks, helping teams simulate disasters without disrupting production. Every test becomes a rehearsal for the real thing, reducing panic and improving coordination when it matters most.

Recovery tests should include identity and encryption keys. A restored database is no use if nobody can authenticate to it.

Learn from Real Deployflow Case Studies

Zilch (FinTech)

When Zilch, a fast-growing BNPL company, needed to scale its payments infrastructure, Deployflow implemented Terraform-based automation on AWS, cutting environment setup time by half and enabling complex API integrations within just one month. 

The result was faster releases, higher platform stability, and the ability to adapt instantly to vendor and security changes, a prime example of how DevOps-driven automation supports cloud resilience and business continuity.

Little Journey (HealthTech)

Deployflow helped Little Journey replace manual, inconsistent infrastructure with automated provisioning using Terraform and Azure DevOps. The result was an 80 per cent reduction in deployment time and 50 per cent higher scalability, proving that automation and secure multi-environment delivery can transform compliance-heavy sectors.

Strike (PropTech)

When property platform Strike lost its internal DevOps team, Deployflow rebuilt its AWS environment from the ground up. By redesigning pipelines, eliminating database outages, and improving observability, Strike achieved a 70 per cent improvement in cloud stability and a 60 per cent reduction in downtime, showing how professional DevOps support turns outages into opportunities.

Vertical infographic highlighting Deployflow case studies for Zilch, Little Journey, and Strike showing improved cloud stability and automation.

What the Three Projects Have in Common

Each project replaced manual, person-dependent infrastructure with automated, version-controlled environments. Every resilience pattern in this article depends on that step, because a region can only be rebuilt in minutes if its infrastructure already exists as code.

Businesses that embrace this mindset don’t wait for the next outage to test their limits; they engineer confidence into every deployment.

Prepare for the Next AWS Outage

AWS will fail again, and the open question is which layer breaks next. Map which failure domain each tier-1 service survives today, then test that answer before an outage does.

Partner with Experts Who Engineer Stability

Outages happen. What matters is how fast you recover. 

Deployflow’s cloud migration, automation, and DevOps managed services help businesses build systems that stay online, even when the internet doesn’t. From multi-cloud architecture design to disaster-recovery orchestration, their engineers embed reliability into every line of code.

With sprint-based full-stack squads, Deployflow delivers progress in fast, measurable iterations so resilience grows with every sprint.

When stability matters most, contact Deployflow and turn uncertainty into uptime.

Frequently Asked Questions About the AWS Outage and Cloud Resilience

How long do AWS outages usually last?

Most AWS incidents resolve within an hour or two. The two largest recent events ran far longer: the October 2025 DynamoDB DNS failure lasted about 15 hours, and the May 2026 thermal event left one Availability Zone impaired for more than a day.

AWS’s 99.99% monthly EC2 target allows roughly four minutes of downtime a month, so a single major incident can exceed a year’s budget.

What causes most AWS outages?

Software and automation faults in the control plane, DNS or networking cause most large AWS outages, including October 2025. Physical failures still happen: the May 2026 outage began with overheating in a single data centre. Automation spreads both kinds quickly, because one faulty update or dependency can cascade across thousands of services.

Outages often occur when multiple internal systems interact in unexpected ways, such as scaling logic exhausting resources or miscommunication between health checks and routing layers.

How can my business prepare for the next AWS outage?

Start by auditing regional dependencies and building redundancy into your infrastructure. Deploy workloads in at least two AWS regions or across multiple providers, automate failover, and regularly test backup restorations.

A structured cloud resilience strategy should include:

  • Multi-region deployments for all tier-1 services.
  • Automated DNS failover to keep traffic flowing.
  • Multi-cloud backup plans to maintain operations if AWS goes dark.

If you lack internal DevOps resources, consider external experts like Deployflow, which specialises in multi-cloud architecture, automation, and disaster-recovery orchestration.

How often does AWS experience downtime?

While AWS rarely faces full regional outages, service interruptions occur multiple times each year. Public data from AWS Health Dashboard shows dozens of localised disruptions annually, usually short-lived but highly visible when they affect high-traffic services.

Because AWS accounts for about 26% of global cloud infrastructure spending, even short incidents can appear massive in scope. This is why continuous monitoring and failover testing are critical; the bigger the provider, the wider the ripple effect of each minute offline.

What’s the difference between multi-cloud and hybrid-cloud resilience?

A multi-cloud strategy distributes workloads across several public cloud providers to reduce dependency on any single vendor. A hybrid-cloud strategy combines public cloud with private or on-premises infrastructure, often to meet compliance or data sovereignty needs. 

Multi-cloud focuses on provider redundancy, while hybrid-cloud ensures continuity even if internet access or a provider connection is disrupted. Both strengthen business continuity when designed with clear recovery processes.

Is AWS down right now?

Check the AWS Health Dashboard, which shows open incidents by service and Region. Signed-in account holders get a personal view in the AWS Health console, which lists only the issues affecting their own resources and notes the AZ ID involved.