Writing · March 10, 2026

AWS vs Azure vs GCP: An engineer's honest comparison

After 7 years deploying production workloads on all three clouds for clients like H&M, Siemens, and Nokia — here's my honest take on when to use which.

I've shipped production infrastructure on AWS, Azure, and GCP. Not tutorials — real enterprise workloads for H&M, Siemens, Nokia, and Alight Solutions. Here's my honest, opinionated comparison based on what I've actually experienced.

The honest summary upfront

• AWS — biggest service catalogue, best documentation, most mature ecosystem
• Azure — best for Microsoft-heavy enterprises, best Active Directory integration
• GCP — best data and ML tooling, most developer-friendly APIs

None of them is universally better. The right choice depends on your workload, your team, and your existing vendor relationships.

Developer experience

AWS has the most services but also the most footguns. IAM alone has a learning curve that takes months to master. The console is dense and overwhelming for newcomers. But the documentation is exceptional — the AWS whitepapers are some of the best technical writing in the industry.

Azure has improved dramatically since 2020. The portal is cleaner than it used to be. The Azure CLI is genuinely good. But the naming conventions are inconsistent (Resource Groups vs Subscriptions vs Management Groups) and the permission model between Azure AD roles and Azure RBAC confuses even experienced engineers.

GCP has the cleanest APIs and the best CLI ergonomics. gcloud is a joy to use. The console is uncluttered. But the service catalogue is smaller and some services feel less production-ready than their AWS or Azure equivalents.

Networking

AWS networking is powerful but complex. VPCs, subnets, route tables, NACLs, security groups, Transit Gateways, PrivateLink — there are many layers and they interact in non-obvious ways. Once you understand it, it's extremely flexible.

Azure networking is more approachable. Hub-and-spoke with VNet peering is well-documented and the tooling around it (Azure Firewall, Private Link, Application Gateway) is solid. For the H&M migration, Azure networking was a strength — setting up 100+ server landing zones was faster than equivalent AWS work would have been.

GCP networking is genuinely different — global VPCs that span regions without peering. This simplifies some things dramatically but requires a mental model shift from AWS/Azure engineers.

Containers and Kubernetes

All three have managed Kubernetes — EKS, AKS, GKE. In my experience:

GKE is technically the most polished Kubernetes experience. AKS is the right choice if you're already in Azure. EKS has the most community resources and integrations.

AI and ML infrastructure

This is where the gap is most pronounced in 2026.

AWS has the broadest AI service portfolio — Bedrock for foundation models, SageMaker for ML pipelines, Textract, Rekognition, Comprehend, Polly, Translate. The breadth is unmatched. Building ReRhythm on Bedrock was a great experience — the API is clean and the model selection is excellent.

Azure has the best enterprise AI story through Azure OpenAI Service — the same GPT-4 models available in ChatGPT, but with enterprise data privacy guarantees and Private Link support. For enterprises that need AI but can't send data to a public API, Azure OpenAI is the answer.

GCP has Vertex AI, which is powerful but has a steeper learning curve. BigQuery ML is excellent for data teams who want to train models without leaving SQL. The Gemini integration across GCP services is improving quickly.

Cost

Nobody wins on cost — it depends entirely on workload. But some observations:

• AWS data egress costs are high. Moving data out of AWS is expensive.
• Azure dev/test pricing is genuinely good for Microsoft shop customers with EA agreements.
• GCP sustained use discounts apply automatically — no commitment required. This surprised me on the Südzucker project.

My actual recommendation

Choose AWS if: you're greenfield, your team is cloud-agnostic, or you need the widest service selection.

Choose Azure if: you're a Microsoft shop (Office 365, Active Directory, .NET), you need Azure OpenAI, or you have an EA agreement.

Choose GCP if: your primary workload is data engineering, ML, or analytics, or your team values developer experience above all else.

Choose all three if: you're me, and clients keep asking.

Working through a cloud provider decision? I'm happy to talk through it on LinkedIn.

Comments & Reactions