It's always DNS.. isn't it?

Over the past month, all major cloud services have gone down - AWS, Azure, Cloudflare... what does AI have to do with it?

GitHub went down again yesterday. For those who aren't familiar, it's basically a social network for programmers - a place to store, share, and collaborate on code.

And over the past month, it feels like we've been watching a full-on marathon of outages across critical cloud services. For example:

1️⃣ The biggest one was likely the AWS outage on 20 October 2025, caused by an error in the DNS system. The service was down for 3 hours and only fully recovered after 9. The damages? Tens - maybe even hundreds - of billions of dollars. During the outage, AWS received 6.5 million reports from dependent services failing: Snapchat, Disney+, Roblox, United and Delta banks, even HMRC.

AWS is Amazon's cloud platform offering virtual servers, databases, and a ton of other services on subscription. There's probably not a single major service on the internet that doesn't rely on AWS in one way or another. It accounts for about 30% of all cloud traffic on the planet.

2️⃣ Then on 29 October, Azure went down - Microsoft's competitor to AWS, holding around 20% of the cloud market. Also due to DNS issues. It was offline for 8 hours. Microsoft 365 and Xbox Live were affected, and Starbucks and various airlines reported disruptions. Estimated losses: between $5 and $16 billion.

3️⃣ On 18 November, it was Cloudflare's turn - the service responsible for website protection, caching, encryption, DNS management, and much more. They handle over 80 million requests per second. Every fifth website uses Cloudflare (mine included). It took them about 6 hours to restore everything. X (formerly Twitter), ChatGPT, Spotify, government portals, airline and railway systems - all affected. They stabilised the situation within 3 hours.

And what's interesting is that in two out of these three cases, the root cause was DNS - arguably the biggest “database” on Earth, translating human-readable website names (like mandrii.com) into the IP addresses computers understand (like 104.21.90.81).

Official explanations attribute this to automation problems and human error... but the frequency and scale of these outages make you wonder: is something more systemic going on?

Has anyone else noticed that over the past several months - or even the past year - software has been just a bit... unstable?

Uncontrolled AI creeping in?