Layer 2 · head to head
Google TPU v7 (Ironwood) vs AWS Trainium 3
Two chips you cannot buy. This is a cloud commitment dressed as a hardware comparison.
| Google TPU v7 (Ironwood) | AWS Trainium 3 | |
|---|---|---|
| Can you get it | Cannot be bought. Exists only inside one cloud, so choosing it is choosing that cloud. | Cannot be bought. Exists only inside one cloud, so choosing it is choosing that cloud. |
| Memory | 192 GB HBM3E per chip | 144 GB HBM3e per chip |
| Bandwidth | 7.37 TB/s per chip | 4.9 TB/s per chip |
| Compute | 4,614 FP8 TFLOPS per chip | 2.52 PFLOPS FP8 per chip |
| Power | Not published | Not published |
| Interconnect | 9.6 Tb/s inter-chip (ICI) | NeuronLink |
| Software | JAX, XLA, PyTorch/XLA | AWS Neuron SDK |
| Workload | inference | both |
Google TPU v7 (Ironwood)
For: Inference at scale on Google Cloud. Ships in 256-chip and 9,216-chip configurations.
The catch: You cannot buy this. It exists only inside Google Cloud, so choosing it is choosing a cloud, permanently. It is also inference-optimised — do not benchmark it as a training part.
Economics: Vertically integrated: the price you see is a cloud price, not a chip price, and it is not comparable to a $/GPU-hour rental line.
AWS Trainium 3
For: Dense training and inference on AWS. Unlike TPU v7 it is designed for both, which makes it the more honest Trainium-vs-TPU comparison.
The catch: AWS-only, and the Neuron SDK is a third software ecosystem to support alongside CUDA and ROCm. Model coverage is the question to ask, not FLOPS.
Economics: Independent analysis has singled out Trainium as the NVIDIA alternative that actually pays off on cost per token — largely because AWS prices it to move.
Before either — can you power it?
Not published against Not published. In 2026 that comparison usually matters more than the FLOPS one: the US interconnection queue exceeds 2,600 GW with waits approaching five years, and roughly 80% of projects withdraw before energising.
Want this whole layer as one table?
The accelerators on these pages as a single CSV — memory, bandwidth, compute, power, interconnect, software stack, availability, who each part is for and the catch — ready to sort, filter or load into a notebook.
Get the silicon table — free for research and evaluation
Four short lines and the download starts at once (CC-BY-NC-4.0, built from these pages the moment you ask). What you type is used only to answer the request and is never sold. Prefer Hugging Face? Request it there.
Using it inside a product, a client report or anything you sell? That needs a commercial licence.