80 GB · some unspecified
Market reference and history
Daily collection · Nov 19, 2026
GPU details and history →Compare GPU rental offers, instance costs and price history.
Build Nov 19, 2026 · 20 older quotes retain their observation dates. Prices are posted rates; availability is not checked.
Lowest listed offers follow your filters. The reference price is the broader on-demand market benchmark; its provider count and basis stay visible.
| Explore | ||||
|---|---|---|---|---|
H100 NVIDIA · 80 GB · some unspecified | $3.20/GPU·hr Hyperstack · On-demand | $4.25/GPU·hr Typical — median of 6 specialist clouds | 11 offers from 11 providers · 10 on-demand, 1 serverless Observed Nov 19, 2026 Reference Nov 19, 2026 | GPU details → |
H200 NVIDIA · 141 GB · some unspecified | $3.99/GPU·hr Hyperstack · On-demand | $5.40/GPU·hr Typical — median of 5 specialist clouds | 9 offers from 9 providers · 8 on-demand, 1 serverless Observed Nov 19, 2026 Reference Nov 19, 2026 | GPU details → |
A100 80GB NVIDIA · 80 GB | $1.35/GPU·hr Hyperstack · On-demand | $1.96/GPU·hr Typical — median of 4 specialist clouds | 8 offers from 8 providers · 7 on-demand, 1 serverless Observed Nov 19, 2026 Reference Nov 19, 2026 | GPU details → |
B200 NVIDIA · 192 GB / 180 GB · some unspecified | $6.00/GPU·hr Hyperstack · On-demand | $8.50/GPU·hr Typical — median of 5 specialist clouds | 8 offers from 8 providers · 7 on-demand, 1 serverless Observed Nov 19, 2026 Reference Nov 19, 2026 | GPU details → |
B300 NVIDIA · 288 GB / 268 GB / 263 GB · some unspecified | $7.10/GPU·hr 1 GPU · full instance total not recorded GPU rate only · additional resources may cost extra Modal · Serverless / per-second | $10.50/GPU·hr Typical — median of 4 specialist clouds | 7 offers from 7 providers · 6 on-demand, 1 serverless Older quote · Oct 20, 2026 Reference Nov 19, 2026 | GPU details → |
L40S NVIDIA · 48 GB · some unspecified | $1.55/GPU·hr GPU rate only · additional resources may cost extra Nebius · On-demand | $1.55–$2.08/GPU·hr Typical — range of 2 specialist clouds | 5 offers from 5 providers · 4 on-demand, 1 serverless Older quote · Oct 20, 2026 Reference Oct 20, 2026 – Nov 19, 2026 | GPU details → |
V100 NVIDIA · 16 GB | $0.25/GPU·hr $0.25/instance·hr · 1 GPU Verda · On-demand | $0.25–$0.79/GPU·hr Typical — range of 2 specialist clouds | 4 offers from 4 providers · on-demand only Observed Nov 19, 2026 Reference Nov 19, 2026 | GPU details → |
MI300X AMD · 192 GB | $2.99/GPU·hr Hot Aisle · On-demand | $2.99/GPU·hr Typical — 1 specialist cloud | 2 offers from 2 providers · on-demand only Observed Nov 19, 2026 Reference Nov 19, 2026 | GPU details → |
GB200 NVL72 NVIDIA · 189 GB | $16.00/GPU·hr $64.00/instance·hr · 4 GPUs Oracle OCI · On-demand | $16.00/GPU·hr Big-cloud list price only big cloud | 1 offer from 1 provider · on-demand only Older quote · Oct 14, 2026 Reference Oct 14, 2026 | GPU details → |
MI355X AMD · 288 GB | $8.60/GPU·hr $68.80/instance·hr · 8 GPUs Oracle OCI · On-demand | $8.60/GPU·hr Big-cloud list price only big cloud | 1 offer from 1 provider · on-demand only Older quote · Oct 14, 2026 Reference Oct 14, 2026 | GPU details → |
RTX 4090 NVIDIA · 24 GB | $0.69/GPU·hr 1 GPU · full instance total not recorded GPU rate only · additional resources may cost extra Beam · Serverless / per-second | No reference price No on-demand list price recorded | 1 offer from 1 provider · serverless only Older quote · Oct 20, 2026 | GPU details → |
RTX 5090 NVIDIA · 32 GB | $1.09/GPU·hr 1 GPU · full instance total not recorded GPU rate only · additional resources may cost extra Beam · Serverless / per-second | No reference price No on-demand list price recorded | 1 offer from 1 provider · serverless only Older quote · Oct 20, 2026 | GPU details → |
TPU v5e Google · Offer memory not recorded | $1.20/chip·hr Google Cloud · On-demand | $1.20/chip·hr Big-cloud list price only big cloud | 1 offer from 1 provider · on-demand only Observed Nov 16, 2026 Reference Nov 16, 2026 | GPU details → |
TPU v5p Google · Offer memory not recorded | $4.20/chip·hr Google Cloud · On-demand | $4.20/chip·hr Big-cloud list price only big cloud | 1 offer from 1 provider · on-demand only Observed Nov 16, 2026 Reference Nov 16, 2026 | GPU details → |
TPU v6e Google · Offer memory not recorded | $2.70/chip·hr Google Cloud · On-demand | $2.70/chip·hr Big-cloud list price only big cloud | 1 offer from 1 provider · on-demand only Observed Nov 16, 2026 Reference Nov 16, 2026 | GPU details → |
Trainium2 AWS · Offer memory not recorded | $2.24/chip·hr $35.76/instance·hr · 16 chips AWS (trn2) · Capacity Blocks | No reference price No on-demand list price recorded | 1 offer from 1 provider · Capacity Block only Older quote · Oct 14, 2026 | GPU details → |
80 GB · some unspecified
Daily collection · Nov 19, 2026
GPU details and history →141 GB · some unspecified
Daily collection · Nov 19, 2026
GPU details and history →80 GB
Daily collection · Nov 19, 2026
GPU details and history →192 GB / 180 GB · some unspecified
Daily collection · Nov 19, 2026
GPU details and history →288 GB / 268 GB / 263 GB · some unspecified
Daily collection · Nov 19, 2026
GPU details and history →48 GB · some unspecified
Latest provider observations · includes retained quotes · Nov 19, 2026
GPU details and history →16 GB
Daily collection · Nov 19, 2026
GPU details and history →192 GB
Daily collection · Nov 19, 2026
GPU details and history →189 GB
Reviewed observations · Oct 14, 2026
GPU details and history →288 GB
Reviewed observations · Oct 14, 2026
GPU details and history →24 GB
32 GB
Offer memory not recorded
Reviewed observations · Nov 16, 2026
GPU details and history →Offer memory not recorded
Reviewed observations · Nov 16, 2026
GPU details and history →Offer memory not recorded
Reviewed observations · Nov 16, 2026
GPU details and history →Offer memory not recorded
Save these filters and optionally set a posted-price threshold. Alerts are checked on this page when you visit, using provider quotes observed within 3 days. Estimates and secondary snapshots do not trigger alerts. Capacity remains unverified.
Saved in this browser. No email, push notifications or background monitoring. Build: 2026-11-19.
Manage saved searches, edit thresholds and back up your Watchlist →
Loading saved searches…
Interruptible rates with their own region and observation date. A discount appears only for a verified matching on-demand offer. Spot capacity can be reclaimed at short notice, so these never enter the typical, cheapest or change figures above.
| GPU | Provider | Spot | On-demand | Discount | Source |
|---|---|---|---|---|---|
| B300 | VerdaRegion · Provider regions vary | $6.21/GPU·hr | $12.41Matched offer · Provider regions vary | −50% | Verda instance-types API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:24:01.211 UTC On-demand observed 2026-11-16 05:24:01.211 UTC Spot SKU: verda:B300 268GB 1-GPU instance 1B300.30V; B300 268GB; 30 vCPUs and 255 GB RAM per instance; published per instance-hour, which is per GPU-hour |
| B300 | NebiusPreemptible · lowest region us-north13 regions, $6.79–$13.40gpu-b300-sxm | $6.79/GPU·hr | $9.50Separate regional floor · uk-south1, eu-west2, us-north1 | Not comparable | Nebius spot-price API · Spot observed Nov 19, 2026Observation detailsSpot observed 2026-11-19 04:02:54.705 UTC Spot effective 2026-11-19 03:51:00 UTC On-demand observed 2026-11-16 05:24:00.440 UTC Spot SKU: sku-e00hfg6dlxtt9ji6fqo9u |
| B200 | NebiusPreemptible · lowest region me-west12 regions, $3.65–$4.25gpu-b200-sxm-a | $3.65/GPU·hr | $8.50Separate regional floor · us-central1, me-west1 | Not comparable | Nebius spot-price API · Spot observed Nov 19, 2026Observation detailsSpot observed 2026-11-19 04:02:54.705 UTC Spot effective 2026-11-19 03:31:00 UTC On-demand observed 2026-11-16 05:24:00.440 UTC Spot SKU: sku-e00k58wq19wjszct65fwm |
| B200 | VerdaRegion · Provider regions vary | $4.75/GPU·hr | $9.51Matched offer · Provider regions vary | −50% | Verda instance-types API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:24:01.211 UTC On-demand observed 2026-11-16 05:24:01.211 UTC Spot SKU: verda:B200 180GB 1-GPU instance 1B200.30V; B200 180GB; 30 vCPUs and 170 GB RAM per instance; published per instance-hour, which is per GPU-hour |
| H200 | NebiusPreemptible · lowest region eu-west13 regions, $1.70–$4.76gpu-h200-sxm | $1.70/GPU·hr | $5.40Separate regional floor · eu-north1, eu-north2, eu-west1, us-central1 | Not comparable | Nebius spot-price API · Spot observed Nov 19, 2026Observation detailsSpot observed 2026-11-19 04:02:54.705 UTC Spot effective 2026-11-19 03:36:00 UTC On-demand observed 2026-11-16 05:24:00.440 UTC Spot SKU: sku-e00dh1moiqg9mc5ix6wj2 |
| H200 | VerdaRegion · Provider regions vary | $3.35/GPU·hr | $6.70Matched offer · Provider regions vary | −50% | Verda instance-types API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:24:01.211 UTC On-demand observed 2026-11-16 05:24:01.211 UTC Spot SKU: verda:H200 141GB 1-GPU instance 1H200.141S.44V; H200 141GB; 44 vCPUs and 170 GB RAM per instance; published per instance-hour, which is per GPU-hour |
| H100 | NebiusPreemptible · lowest region eu-north1gpu-h100-sxm | $2.15/GPU·hr | $4.50Separate regional floor · eu-north1 | Not comparable | Nebius spot-price API · Spot observed Nov 19, 2026Observation detailsSpot observed 2026-11-19 04:02:54.705 UTC Spot effective 2026-11-19 01:11:00 UTC On-demand observed 2026-11-16 05:24:00.440 UTC Spot SKU: sku-e00ljkedv5qtbwhikuoct |
| H100 | Azure (ND v5)Region · eastus (cheapest US region) | $2.27/GPU·hr | $12.29Matched offer · eastus (cheapest US region) | −82% | Azure Retail Prices API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:23:59.498 UTC On-demand observed 2026-11-16 05:23:59.498 UTC Spot SKU: Standard_ND96isr_H100_v5 Per GPU within ND96isr H100 v5 (8× H100) |
| H100 | VerdaRegion · Provider regions vary | $2.62/GPU·hr | $5.24Matched offer · Provider regions vary | −50% | Verda instance-types API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:24:01.211 UTC On-demand observed 2026-11-16 05:24:01.211 UTC Spot SKU: verda:H100 80GB 1-GPU instance 1H100.80S.30V; H100 80GB; 30 vCPUs and 120 GB RAM per instance; published per instance-hour, which is per GPU-hour |
| A100 80GB | VerdaRegion · Provider regions vary | $1.19/GPU·hr | $2.39Matched offer · Provider regions vary | −50% | Verda instance-types API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:24:01.211 UTC On-demand observed 2026-11-16 05:24:01.211 UTC Spot SKU: verda:A100 80GB 80GB 1-GPU instance 1A100.22V; A100 80GB; 22 vCPUs and 120 GB RAM per instance; published per instance-hour, which is per GPU-hour |
| L40S | VerdaRegion · Provider regions vary | $1.01/GPU·hr | $2.02Matched offer · Provider regions vary | −50% | Verda instance-types API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:24:01.211 UTC On-demand observed 2026-11-16 05:24:01.211 UTC Spot SKU: verda:L40S 48GB 1-GPU instance 1L40S.20V; L40S 48GB; 20 vCPUs and 60 GB RAM per instance; published per instance-hour, which is per GPU-hour |
| V100 | VerdaRegion · Provider regions vary | $0.12/GPU·hr | $0.24Matched offer · Provider regions vary | −50% | Verda instance-types API · Spot observed Nov 16, 2026Observation detailsSpot observed 2026-11-16 05:24:01.211 UTC On-demand observed 2026-11-16 05:24:01.211 UTC Spot SKU: verda:Tesla V100 16GB 1-GPU instance 1V100.4V; Tesla V100 16GB; 4 vCPUs and 18 GB RAM per instance; published per instance-hour, which is per GPU-hour |
| RTX PRO 6000 | NebiusPreemptible · lowest region us-central13 regions, $1.19–$1.38gpu-rtx6000 | $1.19/GPU·hr | — | Not comparable | Nebius spot-price API · Spot observed Nov 19, 2026Observation detailsSpot observed 2026-11-19 04:02:54.705 UTC Spot effective 2026-11-19 03:11:00 UTC Spot SKU: sku-e00otwkvc6gea6nmvvsda |
Only providers that publish a spot rate appear; AWS spot prices are not collected. Prices in USD per GPU-hour or chip-hour, as each provider lists them on the date shown.
Nebius spot prices change through the day: each row is the lowest regional spot price in one daily read. An on-demand comparison must match that region, platform and GPU-only configuration, and have been observed on the same UTC day. Otherwise the two rates are separate regional floors and no discount is calculated. Observation details retain both timestamps. Spot prices: Nebius. Not licensed for redistribution: Nebius spot prices are in none of the site’s downloads.
Typical is the median on-demand list price of the covered specialist clouds when at least three quote the accelerator. With fewer providers, the actual price or range stays labeled. The reference price falls back to the big-cloud list price where needed.
Current offers retain each provider’s latest known observation and its date. Newer eligible daily collections replace that provider’s earlier reviewed sample; aging alone never restores an older price. Quotes older than three days at build time are labeled, and a recent-only filter is available. A median needs at least three distinct providers, with one vote per provider.
Lowest listed offer follows the active filters and retains its billing model, source and date. The broader market reference uses on-demand rates. GPU pages, provider pages and the quote checker use the same current offer selection. Instance size, region, CPU and memory charges, and commitment terms remain part of each offer. A posted price never confirms available capacity.
Changes use comparable observations near the stated period boundary and show their actual dates and provider coverage. When the providers behind a typical price change, a change marked “same offers” follows only the offers quoted on both of two collection days, at least two per step, chained together, so a provider joining or leaving does not register as a price move; a day that shares too few offers with the days before it is stepped over. A missing comparison means insufficient history or too little overlap. A recorded zero means comparable prices were equal.
Prices are advertised rates in USD per GPU-hour or chip-hour. Memory filters use the actual recorded offer configuration, with GiB converted to GB for filtering. Unknown memory cannot satisfy a minimum. Instance totals require an explicit accelerator count and may exclude separately billed resources. Gaps show missing days or a changed provider population; insufficient history is never treated as a flat price.
The history view starts with current provider observations. Its basis follows an explicit overview basis unless you override it. A single observation is shown as a compact comparison. Reviewed price series and daily collected indexes remain separately selectable. Charts show real dated observations, with gaps and ranges visible. GPU-hour and chip-hour rates, price bases, and billing models stay separate. Market-implied estimates are available on the research forward-market chart.