I have built on AWS and I would build on it again. It is genuinely the most capable cloud platform in existence, and the breadth of what it offers is not marketing.

I have also been called in to reduce cloud bills on projects that had no business being where they were, and to untangle architectures where a small team had accumulated a large estate because each individual decision seemed reasonable at the time.

Both things are true. The question worth answering is where the line sits.

What AWS Genuinely Does Better

No other provider matches these. If your requirements land here, the decision is made.

Breadth of services. Nearly three hundred of them. When you need a managed streaming platform, a GPU cluster for model training, satellite imagery processing, or a compliance-certified environment for a specific regulator, AWS has it and the alternatives may not.

Regulatory and certification coverage. AWS holds more certifications in more jurisdictions than any competitor. For healthcare, government, and financial services this is often the deciding factor rather than a technical preference.

Enterprise integration. Identity policies, private networking, organisational guardrails, and the depth of access control. For a large organisation with an existing identity provider and compliance programme, AWS integrates into that structure.

Maturity under stress. AWS has been through scale events that most providers have not. When something goes wrong at three in the morning, documentation and community knowledge are deep.

What This Costs

Capability is not free, and the price is paid in three specific currencies.

Operational complexity. AWS is not a platform you deploy to, it is a set of primitives you assemble. A production-grade setup involves network design, subnet tiers, security groups, load balancing, scaling policies, identity roles, log aggregation, and monitoring. Each is correct on its own. Together they are a system that needs an owner.

Cost opacity. The pricing model is granular and the interaction between services is not always obvious. A data transfer choice, a gateway, an unoptimised egress pattern. These produce bills that surprise people, and the surprise usually arrives after the architecture is set.

The knowledge requirement. A team of two generalist developers can operate a simple platform. They cannot comfortably operate a well-architected AWS environment without dedicated attention, and that attention competes with building the product.

What you need AWS reality Simpler platform reality
Live URL in a week Possible, with care Default
Deploy on push Configure a pipeline Built in
TLS and a CDN Multiple services to wire Automatic
Database Managed, and tunable Managed, fewer knobs
Cost predictability Requires monitoring and alerts Usually flat
Person-hours to operate Substantial Minimal

The Pattern We Keep Seeing

The most common failure is not a badly built AWS environment. It is a well-built one that is wrong for the business.

A company builds a product on AWS because AWS is what serious teams use. The architecture is sound. Then the product finds its market, and the founders discover that a meaningful share of their engineering capacity goes into infrastructure rather than product. Not because AWS is bad, but because infrastructure is a genuine discipline and it is competing with the thing that actually generates revenue.

There is also a subtler cost. On AWS a great many things are possible, which means architectural decisions stay open longer. On a constrained platform, the platform makes some decisions for you, and that forced decision-making is often worth more than the flexibility given up.

When AWS Is Obviously Correct

I want to be clear that this is not a case for avoiding AWS.

Choose it when you need services nobody else has, when you operate under a compliance regime that requires its certifications, when you have an infrastructure function that enjoys owning the environment, or when your scale makes the economics favourable. Committed spend, reserved capacity, and savings plans do change the maths at volume.

Also choose it when your data has to be in a specific jurisdiction with specific guarantees, or when you are integrating with an existing enterprise estate that already lives there.

That is a real list, and if you are on it, AWS is the answer.

When It Is Just Overhead

Avoid AWS when the site is a marketing site, an internal tool, a content platform, or an early product still searching for a market. In those cases an edge platform gives you a global network, automatic transport security, instant deploys, near-zero operations, and a bill you can predict on one line.

We serve client sites from an edge network. There is no server to patch, no scaling policy to tune, no egress surprise. A deploy takes seconds. The person who maintains it does not need to be an infrastructure specialist.

Is it as capable as AWS? No. Is that capability needed? For these projects, no.

A Decision Framework

Situation Recommendation
Need a niche managed service only AWS offers AWS
Regulated industry requiring specific certification AWS
Enterprise with an existing AWS estate AWS
You have an infrastructure team AWS is viable
Marketing or brand site Edge platform
Internal tool, small team Edge platform
Early product, unclear requirements Edge platform first
Predictable cost is essential Edge platform

Migrating Is Not Free

One caution. Moving off AWS later is more expensive than starting elsewhere, because architecture accumulates assumptions: network layout, identity structure, service coupling. If there is a real chance the product grows into needing AWS, it is worth designing so that the data layer stays portable rather than deeply coupled.

That said, the honest majority case is that a project starts on AWS for familiarity and never uses the capability it is paying for in complexity. Separating “we could use AWS” from “we benefit from AWS” is the whole exercise.

If you are weighing up infrastructure choices, the question I would ask first is not which provider, but how much attention you are willing to permanently allocate to infrastructure. That number usually settles the decision.