CoreWeave is built for enterprise-scale AI training, with InfiniBand-connected multi-node GPU clusters and Kubernetes-native infrastructure. Thunder Compute is meant for individual developers and small teams who need affordable, on-demand GPU access with minimal setup.
Key takeaways:
- CoreWeave H100 HGX node: $49.24/hr for 8 GPUs ($6.16 per GPU-hr).
- Thunder Compute H100 PCIe: $2.19/hr. A100 80GB: $1.09/hr. Per-minute billing.
- CoreWeave is optimized for 8+ GPU distributed training with InfiniBand. Thunder Compute is for single-node development and fine-tuning.
- Thunder Compute includes 100GB per GPU, VS Code/Cursor/Windsurf extensions, and no egress fees.
What CoreWeave Does
CoreWeave is a GPU cloud provider listed on NASDAQ (CRWV) that targets enterprise AI labs and large-scale training workloads. Its infrastructure centers on bare-metal NVIDIA GPU nodes connected via InfiniBand, with Kubernetes-native orchestration and support for clusters of hundreds of GPUs.
CoreWeave's on-demand H100 pricing is structured around 8-GPU HGX nodes: $49.24/hr per node, normalizing to $6.16/GPU-hr. The A100 8-GPU node runs $21.60/hr, or $2.70/GPU-hr. There is no published single-GPU H100 or A100 option. HGX H100 and A100 NVLink configurations are largely committed on client contracts, so on-demand availability can be limited. Reserved capacity offers discounts of 25 to 60 percent depending on the term.
CoreWeave also holds SOC 2, HIPAA, and GDPR certifications, and operates across North America and Europe.
What Thunder Compute Does
Thunder Compute provides on-demand GPU instances with a focus on low, stable pricing and developer-first workflows. A100 80GB starts at $1.09/hr and H100 PCIe at $2.19/hr, both billed per minute with no egress fees and no contracts.
Every instance includes 100GB per GPU at no extra charge. Snapshots let you save and restore your full environment before changing GPU configurations. The VS Code, Cursor, and Windsurf extensions connect directly to a running instance, so you can write and run code without SSH setup or CUDA driver installation.
Thunder Compute supports 1x, 2x, 4x, and 8x GPU configurations on a single node. It is not suited to multi-node InfiniBand workloads.
Pricing Comparison
| GPU | Thunder Compute | CoreWeave (on-demand) |
|---|---|---|
| H100 80GB | $2.19/hr (1x-8x) | $6.16/GPU-hr (8x min) 1 |
| A100 80GB | $1.09/hr (1x-8x) | $2.70/GPU-hr (8x min) 2 |
| L40 | $0.79/hr (1x-8x) | $1.25/GPU-hr (8x min) 3 |
| RTX A6000 | $0.35/hr (1x) | Not listed |
| Billing | Per minute | Per second |
| Minimum GPUs | 1 | 8 |
| Contracts | None | Required for most large-cluster configs. |
|
1 CoreWeave H100 HGX node: $49.24/hr ÷ 8 GPUs = $6.16/GPU-hr. 2 CoreWeave A100 HGX node: $21.60/hr ÷ 8 GPUs = $2.70/GPU-hr. 3 CoreWeave L40 node: $10.00/hr ÷ 8 GPUs = $1.25/GPU-hr. Last update: July 14, 2026. | ||
Infrastructure and Target Audience
CoreWeave runs bare-metal GPU nodes connected via NVIDIA InfiniBand, providing high-bandwidth inter-GPU communication for large distributed training workloads. CoreWeave's own benchmarks report 51-52% MFU at 1,024 H100 GPUs, above the 35-45% industry average, and a Mean Time to Failure of 3.66 days versus 0.33 days reported by leading AI labs.
CoreWeave suits teams training 70B+ parameter models, running multi-node jobs, or requiring enterprise SLAs and compliance certifications. The 8-GPU minimum on HGX nodes means you pay for eight GPUs even if your workload needs only one or two.
Thunder Compute targets developers and small teams that want single-GPU access at the lowest available managed rate. Per-minute billing, persistent storage, and one-click IDE integration reduce infrastructure friction to near zero.
Developer Experience
CoreWeave is Kubernetes-native, giving infrastructure teams full control over scheduling, networking, and resource allocation. It suits organizations with DevOps expertise and capacity to manage orchestration. Individual developers without Kubernetes experience will find the platform introduces significant setup overhead.
Thunder Compute requires no Kubernetes experience. Install the VS Code, Cursor, or Windsurf extension, add a payment method, and launch an instance in under a minute. Persistent storage means you do not need to manage data volumes or worry about losing your environment between sessions.
When to Choose Each Provider
Choose CoreWeave is best for:
- Large-scale distributed training on 16+ GPUs.
- Workloads that need InfiniBand inter-node bandwidth.
- Enterprise compliance certifications.
- Organizations with the engineering capacity to manage Kubernetes-based infrastructure.
Choose Thunder Compute for:
- Single-node training or fine-tuning at the lowest managed per-GPU rate
- Per-minute billing, persistent storage, and instant IDE access
- Iterative development where spinning up and down frequently matters.
Last Thoughts on CoreWeave vs Thunder Compute
CoreWeave wins at scale, when inter-node interconnect matters, and when enterprise SLAs are non-negotiable. Thunder Compute wins on cost, simplicity, and developer experience for single-node workloads. Most teams doing development, fine-tuning, and experimentation do not need CoreWeave's multi-node InfiniBand infrastructure, and the $6.16/GPU-hr versus $2.19/hr difference is significant at any meaningful job length.
FAQ
How much does CoreWeave charge for H100 GPUs?
CoreWeave's HGX H100 8-GPU node costs $49.24/hr on-demand, or $6.16/GPU-hr. There is no single-GPU H100 option. Thunder Compute H100 PCIe is $2.19/hr with no 8-GPU minimum.
Is Thunder Compute cheaper than CoreWeave for H100 access?
Yes, both for single-GPU workloads and for 8 GPU clusters. CoreWeave only offers 8x configurations for a total of $49.24/hr, while Thunder Compute H100 PCIe is $2.19/hr for a single GPU and $23.12 for 8.
Does CoreWeave require long-term contracts?
CoreWeave offers on-demand pricing but access often requires approval. Reserved capacity on 6-month to 3-year terms unlocks discounts of 25 to 60 percent.
What is CoreWeave best used for?
Large-scale distributed training on multi-node clusters. CoreWeave offers InfiniBand networking, Kubernetes-native orchestration, and 8-GPU minimum HGX nodes for enterprise AI workloads.
What does Thunder Compute offer that CoreWeave does not?
Single-GPU instances at $1.09/hr for A100 and $2.19/hr for H100, per-minute billing, 100GB per GPU, and VS Code, Cursor, and Windsurf extensions. No contracts and no minimum GPU count.
Which provider is better for a small ML team?
Thunder Compute. CoreWeave's 8-GPU minimum and enterprise onboarding add friction for small teams. Thunder Compute offers instant access, per-minute billing, and IDE integration.
What MFU does CoreWeave achieve on H100 clusters?
CoreWeave's own benchmarks report 51-52% MFU at 1,024 H100 GPUs, higher than the 35-45% industry average. MTTF reached 3.66 days versus 0.33 days reported by leading AI labs.
What compliance certifications does CoreWeave hold?
CoreWeave holds SOC 2, HIPAA, and GDPR certifications. These make it suitable for regulated enterprise workloads where compliance is a hard requirement.
Which platform is easier for individual developers to use?
Thunder Compute is easier for individual developers because it offers built-in VS Code in the browser, requires no Kubernetes knowledge, includes automatic persistent disk snapshots, and allows live hardware swaps at any time. CoreWeave targets ops teams and requires Kubernetes expertise to manage the platform effectively.