CoreWeave

GPU cloud built around large InfiniBand training clusters (NASDAQ: CRWV)

Large clustersInference

CoreWeave runs NVIDIA fleets in North America and Europe with managed Kubernetes (CKS) and Slurm on Kubernetes (SUNK) over SHARP-enabled InfiniBand. It publishes on-demand prices per instance, mostly 8-GPU nodes, from $6.155 per H100 GPU-hour to $10.50 for GB200 NVL72, the only published on-demand GB200 NVL72 rate in our data. It is one of two neoclouds rated Platinum in SemiAnalysis ClusterMAX 3.0.

Best for: Large multi-node training on InfiniBand, and on-demand GB200 NVL72.

ClusterMAX 3.0: Platinum

✓ Publishes prices
✓ Serverless inference
– AMD GPUs
– European data centers

Visit CoreWeave

Facts checked November 9, 2026 against the sources listed below. CoreWeave does not pay to appear here.

1 Legal nameCoreWeave

2 HeadquartersLivingston, New Jersey, US

3 Founded2017

4 Funding or listingPublic (NASDAQ: CRWV). Nvidia invested $2B in equity in January 2026.

5 How you rentOn-demand, Reserved clusters, Bare metal, Serverless inference

6 Minimum commitmentNone stated for on-demand (billed hourly); committed-use discounts of up to 60% via sales

7 InterconnectNVIDIA Quantum InfiniBand with SHARP; more than 3.2 Tbps per 8-GPU HGX H100 node

8 OrchestrationKubernetes (managed, CKS), Slurm (SUNK, Slurm on Kubernetes)

9 GPUs offeredGB300 NVL72, GB200 NVL72, B300, B200, RTX PRO 6000, GH200, H200, H100 SXM, L40S, L40, A100 80GB, Vera Rubin (product page)

10 RegionsNorth America, Europe

11 CertificationsSOC 2 (type not stated), ISO 27001, ISO 27017, ISO 27018

12 Customers and partnersSays 9 of the 10 largest foundation-model providers use it (IR boilerplate); homepage shows OpenAI, Meta, Mistral, Cohere, IBM and Google. Meta’s commitments total $35B through 2032 (The Next Web, Jun 2026).

Published prices

On-demand list prices per GPU-hour, as published on November 9, 2026. Reserved and negotiated prices are usually lower.
GPU	Price	Basis	Source
L40	$1.25	on-demand, per GPU; $10.00/hr 8-GPU instance	Pricing page
L40S	$2.25	on-demand, per GPU; $18.00/hr 8-GPU instance	Pricing page
RTX PRO 6000	$2.50	on-demand, per GPU; $20.00/hr 8-GPU instance (High Memory)	Pricing page
A100 80GB	$2.70	on-demand, per GPU; $21.60/hr 8-GPU instance	Pricing page
H100 SXM	$6.155	on-demand, per GPU; $49.24/hr 8-GPU HGX H100 instance	Pricing page
H200	$6.305	on-demand, per GPU; $50.44/hr 8-GPU instance	Pricing page
GH200	$6.50	on-demand, 1-GPU instance	Pricing page
B200	$8.60	on-demand, per GPU; $68.80/hr 8-GPU instance	Pricing page
GB200 NVL72	$10.50	on-demand, per GPU; $42.00/hr 4-GPU instance	Pricing page

Where we recommend them

Best for large training clusters, an alternative.
Best for B200 and GB200 access, an alternative.

Worth knowing before you sign

Debt principal was $35.55B at 2026-06-30; Q2 2026 net loss was $626M on $2.575B revenue, with $640M of net interest expense (10-Q).
Customer concentration: one customer was 36% of Q2 2026 revenue, a second 26% and a third 10% (10-Q).
On 2026-10-14 its 2032 notes reportedly yielded about 15% and the stock fell about 15% in one day (secondary report citing Bloomberg/FT).
Highest published on-demand H100 rate in our data ($6.155 per GPU-hour); committed-use prices are sales-only.
GB300 NVL72 and on-demand B300 are contact sales. Q3 2026 results are due 2026-11-10.

Go to CoreWeave's site

Sources

Work at CoreWeave? Send corrections with a source to editors@bestneoclouds.com. We fix verified errors within five business days.

Compare with

Nebius Full-stack AI cloud with published on-demand GPU prices (NASDAQ: NBIS)
Lambda GPU cloud with self-serve instances and 1-Click Clusters
Crusoe AI cloud with self-serve NVIDIA and AMD GPUs on its own data center campuses
Firmus Asia-Pacific AI cloud and AI factory builder (formerly Sustainable Metal Cloud)
Nscale European AI cloud selling GB300 and B200 capacity (S-1 filed for NYSE: NSCL)