Tier 3 - ML Frameworks and Infrastructure
The bottom tier of the AWS ML stack — maximum control, maximum responsibility.
| Item | What it is | Choose when |
|---|---|---|
| ML Frameworks | Run PyTorch/TensorFlow/etc. yourself | Custom research, specific versions, unusual architectures |
| AWS ML Infrastructure | Compute/silicon underneath (Trainium, Inferentia, GPUs) | You need purpose-built training/inference hardware |
- Most teams should start at Amazon SageMaker AI (Tier 2) and drop here only when they must.