Cheapest
Lowest true cost per GPU-hour, in stock or stock unknown, newest reading per provider
Verda1× · Finland · Iceland$13.05
Fireworks AI1× · 21:13 UTC$20.00
Rented today by 3 providers with a current on-demand offer, from $6.50 to $20.00 per GPU-hour, median $13.05. Every number below links to the page it was read from, with its read date.
Specifications: manufacturer figures.
Cheapest, median and spread among current offers. Medians count one floor per marketplace and need at least three offers.
Today's podium · on-demand · readings of the last 45 days · reference Nov 1, 2026 21:35 UTCCheapest true cost, best throughput per dollar, highest Trust Score. One provider per step, ten places, every row dated.







Current offers ranked by true cost, each with its read date. Older readings stay listed below, in grey, and never enter a ranking.
| Provider | Configuration | Region | Stock | List | True cost | Read | |
|---|---|---|---|---|---|---|---|
| 72× GB300 | — | Stock unknown | $6.50 | $6.50 | 21:13 UTCfresh | Receipt | |
| 1× GB300 SXM32 vCPU · 225 GB RAM | Finland · Iceland | Stock unknown | $13.05 | $13.05 | 21:09 UTCfresh | Receipt | |
| 4× GB300 SXM128 vCPU · 900 GB RAM | Finland · Iceland | Stock unknown | $13.05 | $13.05 | 21:09 UTCfresh | Receipt | |
| 2× GB300 SXM64 vCPU · 450 GB RAM | Finland · Iceland | Stock unknown | $13.06 | $13.06 | 21:09 UTCfresh | Receipt | |
| 1× GB300 | — | Stock unknown | $20.00 | $20.00 | 21:13 UTCfresh | Receipt |
The same hour, divided by what the card gives you; then the smallest real configuration for the most downloaded open models.
| Unit | At the cheapest | At the median |
|---|---|---|
| per GB of VRAM · hour288 GB | $0.023 | $0.045 |
| per PFLOP FP16 · hour2,500 TFLOPS | $2.600 | $5.220 |
| per TB/s of bandwidth · hour8.00 TB/s | $0.813 | $1.631 |
| Model | Needs | Smallest config | $ / month |
|---|---|---|---|
| all MiniLM L6 v2Sentence Transformers | 1 GBFP8 | 1× GB300at Verda | $9,527 |
| paraphrase multilingual MiniLM L12 v2Sentence Transformers | 1 GBFP8 | 1× GB300at Verda | $9,527 |
| bge-m3BAAI | 1 GBFP8 | 1× GB300at Verda | $9,527 |
| Qwen3 0.6BQwen | 1 GBFP8 | 1× GB300at Verda | $9,527 |
| all-mpnet-base-v2Sentence Transformers | 1 GBFP8 | 1× GB300at Verda | $9,527 |
| bge-reranker-v2-m3BAAI | 1 GBFP8 | 1× GB300at Verda | $9,527 |
Today the GB300 rents from $6.50 per GPU-hour on-demand, with a median of $13.05 across 3 providers, each counted once at its cheapest current offer (5 current offers in all, readings of the last 45 days). Prices depend on the provider, the number of GPUs in the machine, the region and whether you rent on-demand, spot or reserved; true cost adds the mandatory add-ons each provider publishes.
3 providers have a current GB300 offer in stock. Cheapest first: Limestone Networks ($6.50), Datacrunch ($13.05), Fireworks Ai ($20.00). Older readings from other providers stay listed on this page, in grey, and are never ranked.
288 GB of HBM3 at 8.00 TB/s. At FP8 that fits all MiniLM L6 v2 (1 GB), paraphrase multilingual MiniLM L12 v2 (1 GB), bge-m3 (1 GB), Qwen3 0.6B (1 GB), all-mpnet-base-v2 (1 GB) on a single card; larger models need two to eight GPUs or a different card.
The lowest true cost among eligible offers is $6.50 per GPU-hour at Limestone Networks for a 72-GPU machine, read within the ranking window. Offers marked unavailable, older than 45 days, superseded by a newer reading, resold or outside the outlier band are listed but not ranked.
Spot (interruptible) capacity starts at $6.53 per GPU-hour; reserved terms start at $5.26. Spot suits jobs that checkpoint; reserved suits machines that run all month.