Prime Intellect

GPU compute marketplace with its own clusters and post-training tools

MarketplaceInference

Prime Intellect aggregates on-demand GPUs from multiple providers (1 to 256 GPUs by the hour, no contract) and quotes dedicated InfiniBand clusters within 24 hours, run with Slurm or Kubernetes. Its compute page lists H100 at $3.50, H200 at $5.48 and B300 at $6.64 per GPU-hour. SemiAnalysis rated it Bronze in ClusterMAX 3.0 and called it the best marketplace tested.

Best for: Marketplace prices with managed Slurm or Kubernetes clusters.

ClusterMAX 3.0: Bronze

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

Visit Prime Intellect

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

1 Legal namePrime Intellect

2 HeadquartersNot disclosed

3 FoundedNot disclosed

4 Funding or listingPrivate. $130M Series A led by Radical Ventures (Jul 2026); over $150M raised in total.

5 How you rentOn-demand, Reserved clusters, Marketplace, Serverless inference

6 Minimum commitmentNone: 1 to 256 GPUs by the hour; dedicated cluster requests need no commitment

7 InterconnectInfiniBand across nodes; SemiAnalysis tested B300 clusters on ConnectX-8 NICs

8 OrchestrationSlurm, Kubernetes, API

9 GPUs offeredH100, H200, B300

10 RegionsNot disclosed

11 CertificationsNot disclosed

12 Customers and partnersClaims Ramp, Zapier and NVIDIA case studies and over 6,000 customers; logos include Goodfire, Arcee and Browserbase.

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
H100	$3.50	marketplace, on-demand price shown on compute page; form factor not stated	Pricing page
H200	$5.48	marketplace, on-demand price shown on compute page; offers vary by provider	Pricing page
B300	$6.64	marketplace, on-demand price shown on compute page; offers vary by provider	Pricing page

Where we recommend them

Best marketplace, our pick.

Worth knowing before you sign

SemiAnalysis observed reliability issues during burn-in.
Listed prices come from a page widget; actual offers vary by underlying provider and availability, and the availability API needs a login.
No security certifications are stated.
Compute is one of several products: it acquired Moondream (Oct 2026) and also sells RL/post-training, sandboxes and inference.

Go to Prime Intellect's site

Sources

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

Compare with

CoreWeave GPU cloud built around large InfiniBand training clusters (NASDAQ: CRWV)
Nebius Full-stack AI cloud with published on-demand GPU prices (NASDAQ: NBIS)
Crusoe AI cloud with self-serve NVIDIA and AMD GPUs on its own data center campuses
Nscale European AI cloud selling GB300 and B200 capacity (S-1 filed for NYSE: NSCL)
Together AI AI cloud for GPU clusters and model inference