AI chip · Amazon
AWS Trainium 2
Second-generation Trainium announced re:Invent 2023; 4x training performance of Trn1. Backbone of the Anthropic Project Rainier deal.
On the DEPLOY graph: 3 campuses carry a cited inventory line for it · 1 reference-design cluster include it.
Market position
AWS's flagship training silicon of 2025. Anthropic's Project Rainier (hundreds of thousands of Trainium 2 chips) is the anchor customer. Priced explicitly against H100/H200 on training-cost-per-token; NeuronLink UltraServers of 64 chips sit against NVL72. Trainium 3 is the announced successor.
DEPLOY editorial. What the vendor PDF cannot tell you.
Physical-AI cross-link
Derived from claims already on record (vendor first-party TDP, throughput, price, and the reference-design cluster registry). The arithmetic is shown per row so it can be audited. No estimated inputs.
- Perf per watt
- 2.60 FP8 TFLOPS/W1,300 TFLOPS ÷ 500 W = 2.60
- Cooling class (across reference designs)
- waterObserved across 1 reference-design cluster containing this chip
What fits in 96 GB
Weights-only footprint for public open-weight LLMs. Pure arithmetic: params × bytes/param. Excludes KV cache and activation memory; add ~10-30% headroom for real serving. A model that does not fit at FP16 may still fit at INT8 or INT4 with quality trade-offs. Not a benchmark.
| Model | Params | FP16 | INT8 | INT4 |
|---|---|---|---|---|
| Llama 3.1 8B dense | 8 B | 16 GB ✓ | 8 GB ✓ | 4 GB ✓ |
| Llama 3.1 70B dense | 70 B | 140 GB ✗ | 70 GB ✓ | 35 GB ✓ |
| Llama 3.1 405B dense | 405 B | 810 GB ✗ | 405 GB ✗ | 203 GB ✗ |
| Llama 3.3 70B dense | 70 B | 140 GB ✗ | 70 GB ✓ | 35 GB ✓ |
| DeepSeek V3 MoE (671B total, 37B active per token) | 671 B | 1342 GB ✗ | 671 GB ✗ | 336 GB ✗ |
| DeepSeek R1 MoE (671B total, 37B active per token) | 671 B | 1342 GB ✗ | 671 GB ✗ | 336 GB ✗ |
| Qwen 2.5 7B dense | 7 B | 14 GB ✓ | 7 GB ✓ | 4 GB ✓ |
| Qwen 2.5 72B dense | 72 B | 144 GB ✗ | 72 GB ✓ | 36 GB ✓ |
| Mixtral 8x7B MoE (46.7B total, 12.9B active per token) | 46.7 B | 93 GB ✓ | 47 GB ✓ | 23 GB ✓ |
| Mixtral 8x22B MoE (141B total, 39B active per token) | 141 B | 282 GB ✗ | 141 GB ✗ | 71 GB ✓ |
| Gemma 2 27B dense | 27 B | 54 GB ✓ | 27 GB ✓ | 14 GB ✓ |
| Command R+ dense | 104 B | 208 GB ✗ | 104 GB ✗ | 52 GB ✓ |
| Kimi K2 MoE (1T total, 32B active per token) | 1000 B | 2000 GB ✗ | 1000 GB ✗ | 500 GB ✗ |
Math: FP16 = params × 2 bytes; INT8 = params × 1 byte; INT4 = params × 0.5 bytes. MoE models sum every expert (full weights on disk), not the per-token active subset.
Common questions
Answer-first, sourced. Every claim below traces to a specific field on this page or a cited datasheet. FAQPage schema is emitted so LLM crawlers can lift these verbatim.
How much does AWS Trainium 2 cost?
AWS Trainium 2 launch pricing is not on record in the DEPLOY registry. Vendor list prices are typically published in the vendor's own briefing rather than the datasheet.
How much memory does AWS Trainium 2 have?
96 GB of HBM3 at 2,900 GB/s. Vendor datasheet, first-party.
How much power does one AWS Trainium 2 draw?
500 W TDP (thermal design power) per chip. System-level draw is higher: see the reference-design section for rack-level kW.
Which data centers have AWS Trainium 2?
3 campuses on the DEPLOY registry carry a cited inventory line for this chip. Full list below with per-campus quantity hints where published.
Which open-weight LLMs fit on one AWS Trainium 2?
Of the 13 public open-weight LLMs DEPLOY tracks: 4 fit at FP16, 7 at INT8, 9 at INT4 (weights only, excludes KV cache). The full table is above with per-model math. Sparsity, offloading and multi-GPU serving change the picture; this row is single-chip weights-only.
Key facts
- Class
- Compute SoC (AI accelerator)
- Designer
- Amazon
- Safety-critical?
- No (data-center inference / training)
- Record as of
- 2026-08-25
- Latest cited claim
- 2024-12-03 (across 4 inventory + benchmark rows)
Specifications (10 fields, click to expand)
Vendor datasheet figures (first-party). Dense throughput first; sparse (2:4) in parentheses where NVIDIA quotes it. The exhaustive spec sheet lives on the datasheet URL below the table: this row set covers what buyers actually compare on.
- Process node
- TSMC N5
- TDP
- 500 W
- Memory
- 96 GB HBM3
- Memory bandwidth
- 2,900 GB/s
- BF16 (dense)
- 667 TFLOPS
- FP8 (dense)
- 1,300 TFLOPS
- INT8 (dense)
- 1,300 TOPS
- Form factor
- AWS Neuron 2 (Trn2 UltraServer with 64 chips per unit)
- Announced
- 2024-12-03
- Released
- 2025-01-01
Source: vendor datasheet
Generation
Used in reference designs
The named rack-scale and pod-scale designs buyers actually order that contain this chip.
- ×64AWS Trainium 2 UltraServerultraserver
Compare with
Benchmarks
Published measurements per workload. Vendor datasheet numbers are first-party (verified posture); MLPerf / InferenceX / press results come in when their license terms + workload naming permit.
| Workload | Value | Unit | Source | As of |
|---|---|---|---|---|
| peak tflops fp8 | 1300 | TFLOPS | vendor | 2023-11-28 |
Data centers stocking AWS Trainium 2
Who has the most? →Every campus on the DEPLOY registry with a public claim of AWS Trainium 2 on site. Quantity hints are unstructured because underlying disclosures vary by order of magnitude.
- Amazon 'Bridgefield' GWU Ashburn Campusas of 2024-12-03announcedinferred
AWS Trainium 2 capacity across the Ashburn / GWU Bridgefield campus
AWS deploys Trainium 2 across its US-East-1 flagship region; per-campus counts not published.
- Amazon Project Rainier Data Center (New Carlisle)as of 2024-12-03partially energizedreported
Anthropic's compute (Project Rainier): hundreds of thousands of Trainium 2 chips
- AWS Susquehanna (Cumulus) Nuclear-Coupled Data Center Campusas of 2024-11-01partially energizedreported
Anthropic Project Rainier training cluster — Trainium 2 at scale (960 MW envelope)
Sources
Adoption rows appear as we verify chip-integration claims per model. Every row is a public claim the maker or a tier-1 source has stated; verification tier (verified / reported / inferred) is shown per row. See every chip on record for the full catalog or the Amazon record.