MLOps · 05 of 5

On any cloud

AWS, Azure, Google Cloud, your data centre or an in-region GPU cloud: self-hosted or managed.

amazon-eksaksgoogle-cloudkuberneteskubeflowmlflow
Self-hosted or managed, on the cloud you run
01

The same lifecycle on AWS, Azure and Google Cloud

Self-hosted on the provider's Kubernetes, or on its managed ML services. NGT's Kubeflow reference design covers all three.

AWSAmazon EKS or Amazon SageMaker
  • Self-hosted: EKS with CPU and GPU node groups, Karpenter for scaling
  • Managed: SageMaker Pipelines and endpoints
  • Data on S3; FSx for Lustre where throughput matters
Amazon EKSAmazon SageMakerAmazon S3FSx for LustreKarpenter
Microsoft AzureAKS or Azure Machine Learning
  • Self-hosted: AKS
  • Managed: Azure Machine Learning for training, registry and deployment
  • Data on Blob Storage, images in Azure Container Registry
AKSAzure Machine LearningBlob StorageAzure Container Registry
Google CloudGKE or Vertex AI
  • Self-hosted: GKE, including Autopilot
  • Managed: Vertex AI Pipelines, Training and Endpoints
GKEVertex AI
02

Beyond the hyperscalers

When data may not go to a public cloud region, the same Kubernetes design runs elsewhere.

  • Your data centreKubernetes on your hardware, with storage such as NFS, Ceph or MinIO.
  • In-region GPU cloudFPT Smart Cloud's FPT AI Factory offers GPU Kubernetes in Vietnam and Japan: an ecosystem option NGT builds on.
  • Portable by designOnly basic infrastructure services, so the platform is not tied to one provider.
Where the same design runs
  1. amazon-eksAmazon EKSthe reference build
  2. aksAKS
  3. GKEincluding Autopilot
  4. Kubernetes in your data centreNFS, Ceph or MinIO
  5. FPT AI Factory GPU Kubernetesecosystem option · VN, JP
03

MLOps as self-service in the developer platform

In an internal developer platform, ML can be a golden path like any other.

SourcesAI Modeling & MLOps case study, NGT (Sep 2026)AI-First DevOps Transformation: Services & Solutions, NGT (Oct 2026)NGT Service Offering v3.2 (Jul 2026)Machine Learning Platform with Kubeflow, NGT reference deckPlatform Engineering v1.2, NGT (Sep 2026)Data Platform Ops & DataOps Service Catalog, NGT (Sep 2026)FPT AI Factory service offering, FPT Smart Cloud (Nov 2025)