Maybe, we are wrong about the cloud
When Amazon was launching Amazon S3 in their 2006 blog, they promised a reliable, highly scalable, low-latency data storage service - in their own words.
Last week, AWS had an outage for many hours. Thousands of services and millions of users were affected: people could not incline their beds, APIs timed out, “privacy” mobile apps were unable to work and well, serverless applications turned out to rely on servers…
This is a major concern…
In this cloud era, everyone seems okay with relinquishing control of their services to a third party vendor e.g Amazon, Google, Alibaba.
Software craftsmen are willingly removing their hands off their data.
Developers and system admins are more than happy because they do not have to manually take care of on prem servers.
C-suites in suits are happy to shout about how they are using “the cloud”.
People like Richard Stallman doubted and questioned this approach many years ago.
Here is a Reddit thread from 17 years ago link
But, we did not listen.
What can we do to stop or mitigate the cascading effect of someone else’s automated computer not behaving as intended?
Solutions
There are a number of ways to rethink or rather to implement the known ways about how to build, host and deliver software services to the users of our systems; especially after the AWS outage (or for any other upcoming outages of a major cloud platform).
- Host across multi-region high availability (HA)
For example, by using Kubernetes, you are able to deploy some workers in different regions in e.g us-east-1 and ap-east-2 HAs.
- Embrace hybrid architectures
Not everything needs to live in the cloud. Local-first and hybrid setups; where compute and data live closer to the edge; can drastically reduce dependency on a single provider.
- Decentralize the stack
Decentralized and peer-to-peer infrastructure where devices share resources directly without a central server.
Peers are equally privileged, equipotent participants in the network, forming a peer-to-peer network of nodes.
- More self hosting
With tools like Kubernetes on bare metal boxes, self-hosting isn’t the nightmare it used to be (especially for medium to large enterprise businesses).
The web wasn’t meant to be this centralized. In 2025, when one company’s outage can take down half the internet (literally), it’s a sign that we have drifted far from the original vision of resilience through distribution.
Maybe we are wrong about the cloud — or at least, wrong about how much of it we should rely on.
The dream was infinite scalability; the reality is infinite fragility.
So next time your favorite app 404s because “AWS is investigating elevated error rates,” remember:
Reliability isn’t a service. It’s a philosophy.