AWS Trainium / Inferentia
Amazon's in-house training (Trainium) and inference (Inferentia) silicon
AWS designs its own AI accelerators through its Annapurna Labs unit: Trainium for training and Inferentia for inference, reducing dependence on Nvidia for internal and customer workloads. Trainium2 scaled in 2025 (notably the massive 'Project Rainier' cluster built with Anthropic), with Trainium3 on a 3nm node announced for 2025-2026. Chips are designed with Marvell/Alchip support and fabbed at TSMC; they undercut GPU pricing on EC2 for large-scale training.
Products
Trainium (train), Inferentia (infer)
2025 milestone
Trainium2 'Project Rainier' w/ Anthropic
Next
Trainium3 (3nm)
Design unit
Annapurna Labs; TSMC-fabbed
How it fits the stack
AWS Trainium / Inferentia with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.
AWS Trainium / Inferentia in the AI stack. AWS Trainium / Inferentia with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.
Graph data (text) — 9 entities, 10 relationships
- AWS Trainium / Inferentia —uses→ High-Bandwidth Memory (HBM)
- AWS Trainium / Inferentia —depends on→ MLIR / OpenXLA / StableHLO
- AWS Trainium / Inferentia —depends on→ TSMC (Taiwan Semiconductor Manufacturing Company)
- Amazon Web Services (AWS) —used by→ AWS Trainium / Inferentia
- Anthropic —used by→ AWS Trainium / Inferentia
- Anthropic —uses→ AWS Trainium / Inferentia
- Anthropic capital stack (Amazon/Google) —uses→ AWS Trainium / Inferentia
- AWS Trainium / Inferentia —partners with→ Marvell Technology
- AWS Trainium / Inferentia —competes with→ Nvidia
- AWS Trainium / Inferentia —designs→ Marvell Technology
Depends on ↑ · 3
Feeds ↓ · 4