← TPUs, AI chips & CPU endpoints
Amazon Web Services
AWS Trainium2 (per chip)
Per-chip estimate from trn2 instances (16 chips, 96 GB each). Sample entry; hourly rates were not on the fetched page.
USD per accelerator- or instance-hour
$2.20
Unverified sample
#4 cheapest of 5 in TPUs, AI chips & CPU endpoints
Official pricing ↗Capabilities, explained
What this API can do, and what each capability means for your project.
Accelerator type
trainium
The kind of hardware you are renting.
Memory
96 GB
Memory per chip or instance. Sets the largest model you can load.
Frameworks
pytorch, jax
Software stacks that run on this hardware.
Supports training
Yes
Can be used to train models, not only to serve them.
Per-minute billing
Yes
You pay by the minute instead of rounding up to a full hour.
Price history
Only one price recorded so far. The chart appears once the price changes.