deploybase

GH200 on CoreWeave: Pricing, Specs & How to Rent

Deploybase · October 9, 2025 · GPU Pricing

Contents

GH200 CoreWeave Pricing

Gh200 Coreweave Pricing is the focus of this guide. Single GH200: $6.50/hour. 8x GH200 cluster: $50.44/hour ($6.31 per GPU). Reserve annually, get 15-25% off.

No spot instances. CoreWeave bundles compute, networking, storage:no hidden fees.

Deploys in 5-10 minutes.

GH200 GPU Technical Specifications

The GH200 combines a Grace CPU with an H100 Tensor Core GPU in a unified memory architecture. The Grace processor contains 72 ARM-based cores capable of substantial FP32 performance paired with the H100 GPU.

The integrated design provides 900 GB/s Grace-to-GPU memory bandwidth via NVLink-C2C, approximately 7x higher than PCIe Gen5. This enables fine-grain data movement optimization impossible with discrete GPU systems. Applications can treat Grace CPU memory and GPU memory as unified address space.

H100 components within GH200 deliver 67 teraflops FP32 throughput and 989 teraflops BF16 tensor performance (1,979 TFLOPS with sparsity). Aggregate bandwidth across eight H100 GPUs reaches 26.8 TB/s, enabling massive batch processing and distributed model serving.

Compare H100 specifications for discrete GPU alternatives or explore data center GPU pricing at competing providers.

How to Rent GH200 on CoreWeave

CoreWeave customers access GPU instances through web dashboard or REST API. Account creation requires email verification and payment method registration. Credit card and ACH wire transfers are accepted; cryptocurrency funding is not available.

Provisioning begins with instance configuration. Teams select cluster location (US East, US West, Europe), storage capacity, and networking requirements. Most GH200 instances deploy within 5-10 minutes, significantly faster than Azure's or AWS's sales-cycle-dependent provisioning.

CoreWeave provides SSH access with persistent storage at $0.10/GB per month. Instances support custom container images, enabling Docker-based workflow portability. Dedicated support teams assist with optimization and troubleshooting.

Billing follows hourly metering with automatic monthly invoicing. No minimum monthly commitments required, though annual reserved instances reduce per-hour rates by 15-25%.

Competitive GH200 Provider Analysis

Lambda Labs charges $1.99 per hour for GH200 instances, approximately $4.51 per hour lower than CoreWeave's $6.50. Lambda's aggressive pricing reflects recent GH200 inventory acquisition and market penetration strategy.

RunPod does not currently offer GH200 instances as of March 2026.

Azure lists GH200 through specialized instance families at $4.00-$5.00 per hour, undercutting CoreWeave by $1.50-$2.50. However, Azure's minimum multi-month commitments and complex procurement reduce flexibility compared to CoreWeave's pay-as-developers-go model.

For teams prioritizing cost minimization, Lambda's pricing is compelling. For teams valuing immediate availability and simplicity, CoreWeave's managed service justifies premium pricing. See Lambda GPU pricing for budget alternatives.

GH200 Use Cases on CoreWeave

Inference serving at massive scale uses GH200's unified memory architecture. Deploying 70-405B parameter models requires memory-efficient architectures that GH200's 960GB/s Grace-to-GPU bandwidth enables. Multi-instance serving distributes load across multiple GH200 systems.

Research and development on latest architectures benefits from GH200's Grace CPU capabilities. Academic institutions and research labs explore CPU-GPU co-optimization techniques impossible with discrete systems.

Real-time data processing pipelines use GH200's CPU capabilities. Grace's 72 cores handle preprocessing, feature extraction, and post-processing while H100 GPUs accelerate numerical workloads. Single-system execution reduces latency and complexity.

Large-scale batch processing of multimodal data employs GH200's compute density. Processing billions of text-image pairs, video frames, and audio sequences scales efficiently across multiple GH200 instances.

Advantages and Limitations

CoreWeave excels at infrastructure simplicity. No complex sales processes or multi-week provisioning delays. Developers create accounts, deposit funds, and launch GH200 instances within minutes.

Transparent pricing without hidden fees appeals to cost-conscious teams. CoreWeave publishes rates online without regional premiums or surprise surcharges.

GH200's unified memory architecture provides performance advantages unavailable from discrete systems. Grace CPU capabilities reduce heterogeneous application complexity.

High per-hour costs represent the primary limitation. GH200's $6.50/hr rate accumulates rapidly. A continuous month-long job costs $4,680, comparing unfavorably against H100 alternatives at $2,000-$3,000.

Limited availability outside CoreWeave restricts options. Unlike H100 commoditization, GH200 inventory remains concentrated, reducing provider competition and pricing pressure.

FAQ

Is GH200 on CoreWeave cheaper than alternatives? CoreWeave's $6.50/hr is expensive compared to Lambda's $1.99/hr for GH200 or RunPod's $2.69/hr for H100. CoreWeave's premium reflects immediate availability and managed service. Budget-conscious teams should evaluate Lambda or Azure reserved instances.

Why would teams choose CoreWeave's GH200 over Lambda? CoreWeave offers instant provisioning without commitment. Lambda requires account setup and often has multi-day lead times for capacity allocation. Teams with time-sensitive projects benefit from CoreWeave's speed despite higher costs.

Can I run standard AI workloads on GH200? Yes. GH200's H100 components execute any CUDA code compatible with H100. The Grace CPU provides additional compute capacity and memory bandwidth improvements. Existing H100 workloads migrate to GH200 with minimal modification.

What's the difference between GH200 and H100 for training? GH200's Grace CPU accelerates preprocessing and data loading, reducing overall training time despite identical GPU components. For disk-I/O-bound workloads, GH200 provides measurable advantages. For GPU-compute-bound training, H100 and GH200 perform similarly.

Does CoreWeave offer volume discounts for GH200? Yes. Reserved capacity on annual commitments reduces rates 15-25%. Long-term GH200 commitments may qualify for additional discounts requiring direct sales contact.

Sources