GB200
NVIDIA GB200, Blackwell, 186 GB
- Index
- Not enough live prices for an index yet
- per GPU-hour, As of –
- 7D %
- no data yet
- Providers
- 1
- 38 offers
- Range
- 27.0424.1622.29 – 54.0848.3244.58
- per GPU-hour
The price engine is not reachable right now. Shown: the last data received, as of Oct 25, 2026, 21:05 UTC.
Daily index of GB200 per GPU-hour
Charts start with 14 days of index data. Expected from Nov 8, 2026.
Factsheet
- Index
- Not enough live prices for an index yet
- Cheapest current offer
- USD 27.04EUR 24.16CHF 22.29 Microsoft Azure, United States
- Providers
- 1 (38 offers, 0 in index)
- Countries
- 12
United Arab Emirates
Australia
Brazil
Canada
Germany
Spain
United Kingdom
Indonesia
India
Italy
South Korea
United States
- Price per pricing model: On-demand
- USD 27.04EUR 24.16CHF 22.29 cheapest
- Price per pricing model: Spot
- USD 27.04EUR 24.16CHF 22.29 cheapest
- Price per pricing model: Reserved
- no offer
- High since start
- –
- Low since start
- –
- Range
- 27.0424.1622.29 – 54.0848.3244.58
- 1D
- no data yet
- 30 days
- no data yet
- Year to date
- no data yet
- Pricing models
- On-demand, Spot
- Use cases (manufacturer)
- Training large models, Inference, HPC and simulation Manufacturer page
- As of
- Oct 25, 2026
No index yet: the index needs prices from at least 3 providers. Single offers stay visible.
Cheapest offers
All offers (38)| Provider | Configuration | Location | Price per GPU-hour |
|---|---|---|---|
| Microsoft Azure | GB200 x4 | United States | 27.0424.1622.29 |
About this GPU
GB200 combines a Grace CPU with two Blackwell GPUs on one board. NVIDIA builds it into the rack-scale GB200 NVL72 system for very large inference and training workloads.
GB200 is a data center GPU by NVIDIA based on the Blackwell architecture, introduced in 2024. Offered variants: GB200. Specifications are manufacturer figures.
Source: manufacturer: www.nvidia.com
News
Independent reports
- B300 and GB200 Availability for Training Clusters in EuropeLyceum, Oct 5, 2026
Provider blogs
- CoreWeave Becomes One of the First Cloud Providers to Achieve NVIDIA Exemplar Cloud Validation for Inference on NVIDIA GB200 NVL72Provider blogCoreWeave, Oct 22, 2026
- CoreWeave Sets New Standard as First NVIDIA GB200 Exemplar Cloud, Improving Upon NVIDIA’s Own Training Performance TargetsProvider blogCoreWeave, Oct 22, 2026
Reports that mention this GPU, from the Intelaris news feed. Links go to the source.
Daily index of GB200 per GPU-hour
Charts start with 14 days of index data. Expected from Nov 8, 2026.
Specifications
Manufacturer figures as delivered by the price engine. Tensor values as stated: dense or with sparsity.
| Model | Memory | Type | Bandwidth | FP8 | FP16 | FP32 | Tensor values | Max. power | Interconnect | Source |
|---|---|---|---|---|---|---|---|---|---|---|
| GB200 | 186 GB | HBM3e | 8.00 TB/s | 5,000 | 2,500 | 80 | dense (derived: sparsity value / 2 as stated in footnote) | 1,200 W | NVLink 5; NVLink-C2C to Grace CPU | dam-cdn.nvd.orangelogic.comretrieved Oct 18, 2026 |
FP8, FP16, FP32: TFLOPS.
How these figures are calculated
The index is the typical price of a GPU: the median of the cheapest on-demand price per GPU-hour of each direct provider (hyperscalers excluded), taken from the provider's latest live fetch within the last 3 days, with outliers removed; it is shown only when at least 3 providers have a price.
The cheapest current offer is the single lowest comparable on-demand price of any provider and source, including hyperscalers and published prices up to 14 days old.
Changes over 1, 7 and 30 days and the price chart use the same daily index, and a change compares only providers with a price on both days, so a provider joining or leaving does not move the number.
- index
- median across direct providers (hyperscalers excluded) of each provider's cheapest comparable on-demand price per GPU-hour from its latest live retrieval within the last 3 days (official API, public price page or partner feed), outliers (below 1/3 or above 3x the median) excluded; published only with at least 3 providers
- cheapestCurrentOffer
- lowest current comparable on-demand price per GPU-hour of all providers and sources (live, published within 14 days, hyperscalers included), outliers excluded
- providers
- All providers with a current offer for this GPU (GPU-only partial prices and offers without a stated pricing model excluded), any source and pricing model, hyperscalers included; usually more than indexProviders.
- change
- Change of the index against the last value on or before 1, 7 or 30 days ago, over the providers with prices on both days.
- high52w
- Highest and lowest daily index of the last 52 weeks; until 52 weeks of data exist, counted from dataSince.