NVIDIA
NVIDIA GB200 Grace Blackwell Superchip GPU
Blackwell · 186 GB HBM3E · Large-scale AI training / inference
Integrated in Grace Blackwell systems; 186GB framebuffer appears in NVIDIA MIG documentation.
Loading
Loading workspaces…
NVIDIA GB200 Grace Blackwell Superchip GPU from CoreWeave in North America, Multiple / Unspecified, listed as “NVIDIA GB200 NVL72 · inference platform GPU-hour rate · allocation requires confirmation”.
Recorded commercial and technical data. Reconfirm current price and capacity before procurement.
At a glance
Not yet recorded Configured GPUs · Provider SKU
Original listed price and Kovara’s normalized comparable rate are kept separate.
Comparable rate
$10.50USD / GPU-hour · normalized by Kovara
Listed as USD 10.50 per GPU/hour
Not yet recorded Billing Granularity · Minimum Term · Available From · Setup Fee · Reservation Fee · Minimum Spend · Storage Cost · SLA · Support Tier · Support Model · Deployment Lead Time · Commercial Terms · Support Resource
Take a sourced brief to CoreWeave and confirm there. Kovara does not launch or bill capacity.
Last checked 29 Sept 2026. Price and availability may have changed since; treat the brief as a starting point.
Kovara does not launch, reserve or bill this capacity, and a catalogue record is not a quote or a guarantee of stock. You deploy with CoreWeave under your own account and terms.
# Kovara offering brief
# Generated 2026-11-17T14:20:45.186Z. Values are catalogue records, not a reservation or a quote.
# Reconfirm price and availability with the provider before you commit.
kovara:
offering_id: "OFR-0613"
provider:
id: "PRV0009"
name: "CoreWeave"
url: "https://www.coreweave.com/pricing"
gpu:
id: "GPU0006"
name: "NVIDIA GB200 Grace Blackwell Superchip GPU"
vram_gb: 186
region:
id: "REG0070"
name: "North America"
country: "Multiple / Unspecified"
requested_gpus: null
terms:
deployment_type: "Inference single GPU"
contract_type: "Inference platform"
billing_model: "Hourly"
minimum_term_months: null
price:
original:
value: 10.5
currency: "USD"
unit: "GPU/hour"
normalized_usd_per_gpu_hour: 10.5
availability:
status: "Inference platform customers"
recorded_gpu_count: null
evidence:
confidence_grade: "A"
last_checked: "2026-09-29T05:15:52.064273+00:00"
Yes and No are recorded values. Not verified means Kovara does not hold a reviewed value.
Canonical hardware registry fields for this accelerator.
NVIDIA
Blackwell · 186 GB HBM3E · Large-scale AI training / inference
Integrated in Grace Blackwell systems; 186GB framebuffer appears in NVIDIA MIG documentation.
Provider
CoreWeave is a cloud purpose-built for AI, providing infrastructure and platform services for training, inference and other accelerated workloads.
Not yet recorded Parent Company
Region
Broad/provider-level region descriptor from verified public pricing source; not a single datacenter location. Exact provider label is preserved in offerings.provider_region_name.
Not yet recorded City / Metro · State / Province · Coordinates · Primary Currency · EU / EEA · Data Residency · Electricity Market
No provider data-residency constraint is published for this region.
Compute spend per nameplate GPU kWh: $8.75.
Hourly compute price divided by GPU nameplate power. This comparison excludes host and cooling energy and is not a measured energy cost.2 public price observations · 0 capacity observations
| Observed | Price | Normalized | Type | Confidence |
|---|---|---|---|---|
| 29 Sept 2026, 05:16 UTC | USD 10.50GPU/hour | $10.50 / GPU-hr | Public List Price | Grade A |
| 23 Sept 2026, 00:30 UTC | USD 10.50GPU/hour | $10.50 / GPU-hr | Public List Price | Grade A |
Public evidence supporting this market record. Source reliability is distinct from offering confidence.
A recorded grade does not certify every field or confirm current capacity.
Kovara confirms current price, capacity and terms with the provider before anything is committed.
No verified public capacity observations are currently recorded for this offering.
Where this compute physically sits, at the most precise level supported by public evidence.
No public physical-location relationship. Kovara does not infer an exact building from a Provider region. A facility, campus or metro will appear here only when supported and authorized for publication.
Open Data Center Atlas →