Trainium 3
Trainium 3 is AWS's third-generation custom AI training chip, announced at re:Invent in December 2024.
Text reviewed October 5, 2026
Trainium 3 is AWS's third-generation custom AI training chip, announced at re:Invent in December 2024.
Basic
Announced at re:Invent 2024, Trainium 3 is AWS's custom training chip. Key specs: 5nm process (TSMC), ~860 FP8 TFLOPS per chip, HBM3e memory. Deployed in Ultraclusters of up to 100K chips via EFA networking.
Deep
Trainium 3 targets training cost reduction for frontier models. The chip uses a NeuronCore architecture (custom AWS design, not CUDA-compatible). Software stack is AWS Neuron SDK + PyTorch XLA.
Expert
Neuron compiler supports PyTorch and JAX via XLA; native CUDA code does NOT run.
Depending on why you're here
- ·Amazon's own AI training chip · alternative to NVIDIA
- ·Used to train Claude models on AWS
- ·Access via AWS only · no retail
- ·Requires AWS Neuron SDK + PyTorch XLA
- ·Best for training workloads already on AWS
- ·AWS's counter-bet to NVIDIA dominance · key for AWS margin protection
- ·Gross margin on Trainium inference is much higher than reselling NVIDIA
- ·3rd-gen AWS training ASIC · 5nm TSMC
- ·~860 FP8 TFLOPS · HBM3e
- ·NeuronCore architecture · not CUDA-compatible
The Project Rainier 100K cluster proves the concept · execution is the next question.