The recent Amazon Web Services (AWS) outage on 20 October 2025 demands our urgent attention. When AWS’s US-EAST-1 region experienced DNS resolution failures, it wasn’t just social media that went dark—patient care was directly compromised.
At least ten NHS trusts found themselves in “downtime scenarios,” forced to revert to paper records and business continuity procedures (1). And Central and North West London NHS Foundation Trust temporarily lost power to desktops (1). System C, supporting over 700,000 NHS users, had to invoke emergency protocols (1). Across the Atlantic, United Healthcare’s provider search tool malfunctioned throughout the day (2), whilst mental health clinics in Houston couldn’t validate patient insurance, and radiation therapy scheduling systems became unreliable (3).
This raises fundamental questions for clinical safety: how do we risk-grade infrastructure we don’t control, and more importantly—is cloud architecture appropriate for all healthcare functions? Interestingly, Oracle-Cerner has its own cloud, but not all of it’s services are transferred there yet.
Currently, most healthcare organisations treat cloud providers as low-risk dependencies. Yet AWS commands 32% of global cloud infrastructure (4), creating what regulators might call a “systemic vulnerability.” When a single provider experiences failure, the cascade affects hospitals globally. We’ve seen this pattern before—the 2024 CrowdStrike incident, now this AWS outage. Each incident exposes the same uncomfortable truth: critical clinical systems depend on infrastructure outside our governance.
Perhaps we need to challenge the “cloud-first” orthodoxy. Should real-time clinical systems—those supporting emergency departments, theatre management, or critical care monitoring—rely on internet connectivity and third-party infrastructure? Or, should they be fireproofed, in a way that resilience is built in and a fail-safe-switchover to another server farm be the better approach. These systems demand 99.99% uptime, yet we’re accepting dependencies on services with published SLAs that permit hours of downtime monthly (5).
For non-time-critical functions like data analytics, research repositories, or archival systems, cloud makes economic and technical sense. But for life-critical applications, do we need hybrid architectures with on-premises resilience. Should clinical decision support systems have local fallback capability independent of external connectivity?
If cloud dependency is inevitable, our assurance frameworks must evolve something that ensure healthcare data is prioritised. Should it be all data in one bucket, or several, like a multi-cloud approach with significant redundancy? Do we need to consider geographic resilience? Why is the majority of cloud-based systems based in the US east cost area of Virginia? , surely that’s a risk on several levels. Should there be a UK/European primary farm rather than US, or should the NHS consider having a national redundancy server? While the NHS’ priority is patient care should it now consider a resilience strategy especially as other while less critical services like HMRC were inaccessible, could government services consider now what is needed? I suspect that is a question being asked in Whitehall.
For end users, like nurses, pharmacist’s, doctors and primary and social care, finding out it’s a server issue from the BBC, and not automatically notified by vendors, cannot be the default. Trusts and organisations need to invoke downtime and business continuity procedures usually within an hour, often less in critical areas, the worst thing, and I’ve experienced it personally, is ‘not knowing’ when it will come back up.
As a Clinical Safety Officer (CSO) DCB0129 requires us to assess and mitigate clinical risks, but our risk registers rarely account for hyperscale failure. We classify individual system risks as ” or “low,” or even very low – based on likelihood, yet collectively we’ve created a “high” systemic risk through shared infrastructure dependency in particular as we move more onto cloud-and digitally.
The question isn’t whether to use cloud—that ship has sailed. The question is whether we’re being clinically responsible about which functions we migrate, how we architect resilience, and whether our assurance processes genuinely reflect the risk landscape. When a DNS failure in Virginia can stop patient care in London, Sheffield, and Liverpool, our assurance frameworks aren’t adequate.
It’s time for honest conversation: not to abandon cloud, but to demand better. Better architecture, better assurance, better SLAs, and perhaps, better recognition that not everything belongs in someone else’s data centre and how do we reduce risk. ICB’s should be asking themselves how we get assurance, not through 99.4% uptime reports but what about that 0.6% and its significant impact. What does that mean as most, if not all our data is now digital and inaccessible.
I for one will be reviewing with my clients their hazards associated with cloud hosting, to consider whether the likelihood and consequences are as up to date and what other mitigations we should be considering. It was AWS this week, what could be next?
References
- Digital Health News (2025) AWS outage causes disruption to patient care across NHS sites. Available at: https://www.digitalhealth.net/2025/10/aws-outage-causes-disruption-to-patient-care-across-nhs-sites/ (Accessed: 23 October 2025).
- NBC News (2025) Major AWS outage takes down web services like Snapchat and Ring. Available at: https://www.nbcnews.com/news/us-news/amazon-web-services-outage-websites-offline-rcna238594 (Accessed: 23 October 2025).
- CNN Business (2025) Amazon’s global outage exposes major vulnerabilities to American life. Available at: https://www.cnn.com/2025/10/21/tech/aws-outage-internet-vulnerable (Accessed: 23 October 2025).
- The Nu-Age Group (2025) AWS Outage October 2025: Why Private Cloud Prevents Failures. Available at: https://www.thenuagegroup.us/aws-outage-october-2025/ (Accessed: 23 October 2025).
- Data Centre Magazine (2025) AWS Data Centre Disruption Causes Global Service Outages. Available at: https://datacentremagazine.com/news/aws-down-the-billion-dollar-impact-of-cloud-dependency (Accessed: 23 October 2025).
