01AGPU cloud · accelerator ledger· manual survey 2026-10-14 · automated rows 2026-10-14
Trainium2 rental prices
What capacity costs. Compare the active venue stack for Trainium2, the provider-class spread, and the long-run price direction where history is available.
Trainium2 · market coverage
1 venue
AWS (trn2)
Trainium2 · quote status
Provisional
Evidence observed 2026-10-14
Trainium2 · specs and price per PFLOP-hour
Spec sources on /hardware →Spec-sheet row: Trainium2, reviewed 2026-10-15 against vendor documents; est marks a figure that review did not confirm. Dense tensor rates only, one precision per column — never a sparse figure.
- Dense BF16
- 667TFLOPS
- Dense FP8
- 1,299TFLOPS
- Dense FP4
- not listed on the spec sheet
- Memory
- 96GiB
- Bandwidth
- 2.9TB/s
- TDP
- 500West — estimated / secondary-source figure
- $ / PFLOP·hr · BF16
- No specialist on-demand quote to price it
No specialist on-demand quote to price it, so no price per PFLOP-hour is shown.
Comparable price history
Loading evidence ledger…
Trainium2 · daily venue-class bands
“Big-cloud list price”, “Typical — median of specialist clouds” and “Marketplace floor” per chip-hour, each with its venue count (n)
More actions for Trainium2 · daily venue-class bands
Data — endpoints, license, and freshness behind Trainium2 · daily venue-class bands
The complete GPU price-history ledger: every series and dated observation, with billing, comparison, quality, and raw-evidence fields.
curl -s https://aicomputetracker.com/price-history.jsonPer-accelerator shard of the price-history ledger — the same series and point shapes filtered to one chip.
curl -s https://aicomputetracker.com/price-history/trainium2.jsonNo API key, CORS open (Access-Control-Allow-Origin: *) — fetch it from a browser, a notebook, or a cron job.
The site’s compilation is licensed CC BY 4.0 — attribution for this chart’s data “AI Compute Tracker (aicomputetracker.com), CC BY 4.0”. Third-party source data — such as Kalshi market data, Vast.ai marketplace data, and providers’ list prices — remains subject to its sources’ terms.
Live quotes
Cite — copy a citation for Trainium2 · daily venue-class bands
Copy-ready citation
Cite as: AI Compute Tracker, “Trainium2 · daily venue-class bands”, live quotes as of 2026-10-14, https://aicomputetracker.com/gpu-cloud/trainium2/#class-bands, AI Compute Tracker (aicomputetracker.com), CC BY 4.0, accessed <date you visit>
Replace “<date you visit>” with your access date.
No archived observation days for Trainium2 yet
The daily archive records this chip once a published snapshot carries a comparable quote. The band chart builds as the daily archive grows — it draws once two or more observation days exist.
Daily class aggregates from published Neon quote snapshots, archived through 2026-10-14. A day or class without a comparable quote stays a gap — never interpolated; a median on fewer than 3 venues is drawn as its range (a vertical bar), never as a point on the line, and the tooltip and table view print that range.
Trainium2 · current published quotes
Hourly prices retain each venue’s billing model, cheapest first. Observation dates vary by row: unavailable sources retain their last valid quote, never a new observation date. All rows: Provisional — Dated source-page observation; the raw payload is not yet pinned to the archive.
| Provider | Class | Billing | $ / chip·hr | Notes | Evidence | Observed |
|---|---|---|---|---|---|---|
| AWS (trn2) | Hyperscaler | Capacity Blocks | $2.24 | Published per-accelerator Capacity Block rate; $35.7608/instance-hour. Reservation paid up front; OS fees may apply. | Source — AWS (trn2) published pricing, AWS (trn2) |
Background reading: AWS Trainium 2 on AI Wiki. AI Wiki is run by the same team as AI Compute Tracker.