What it is
Trainium is AWS's family of AI accelerators designed for training and inference workloads that need scalable performance, high memory bandwidth, and cost-aware infrastructure choices.
AWS AI silicon field guide
A compact, developer-focused guide to AWS Trainium accelerators, EC2 Trn infrastructure, and the Neuron software stack for large-scale AI training and inference.
Overview
Trainium is AWS's family of AI accelerators designed for training and inference workloads that need scalable performance, high memory bandwidth, and cost-aware infrastructure choices.
Trainium powers Amazon EC2 Trn instance families and UltraServer configurations, giving teams options from individual accelerated instances to tightly connected large-scale systems.
The AWS Neuron SDK connects model code, compilers, runtime, profiling, and kernel-level optimization so teams can move PyTorch and related workflows onto Trainium infrastructure.
Trainium family
The original Trainium generation powers EC2 Trn1 instances and introduced a dedicated AWS path for deep learning training economics.
Trainium2 powers Trn2 instances and UltraServers, with NeuronLink connecting chips for demanding foundation-model training and inference.
Trainium3 targets agentic, reasoning, video, and multimodal workloads with higher compute, memory capacity, bandwidth, and energy efficiency.
Developer path
Start with familiar ML frameworks, profile the workload, then tune the layers that matter. Trainium adoption is less about a single chip and more about a full stack: instance choice, compiler, runtime, observability, and scaling strategy.
Official references
FAQ
No. Trainium Guide is an independent educational page that links back to official AWS resources for primary product details.
No. AWS positions current Trainium infrastructure for both training and inference across generative AI workloads.
Start with the AWS Trainium overview, then read the Neuron documentation for framework support, compilation, runtime behavior, and profiling workflows.