CoreWeave
GPU cloud infrastructure built for large-scale AI training and inference workloads
CoreWeave operates 32+ data centers with over 250,000 GPUs, purpose-built for AI workloads on a Kubernetes-native architecture. Offers NVIDIA A100, H100, and GB200 GPUs with usage-based, per-minute billing and no data transfer fees. Full 8-GPU HGX nodes start around $49/hr with up to 60% discounts for committed usage. Backed by a $2 billion investment from NVIDIA. Primarily serves enterprise customers for large-scale model training and inference.
Pricing: Usage-based
CoreWeave prices by model
Per 1M tokens, read off CoreWeave's own pricing page on the date shown.
| Model | Input / 1M | Cached / 1M | Output / 1M | Checked | Notes |
|---|---|---|---|---|---|
| MiniMax M3 | $0.23 | $0.05 | $0.96 | 27 Sep 2026 | |
| Qwen3.8 27B | $0.40 | $0.15 | $3.00 | 27 Sep 2026 | |
| gpt-oss-120b | $0.03 | โ | $0.17 | 27 Sep 2026 | |
| gpt-oss-20b | $0.03 | โ | $0.13 | 27 Sep 2026 |
What CoreWeave is
CoreWeave is a GPU cloud built for large-scale AI training and inference. It targets the high end of the market: multi-node clusters of current NVIDIA hardware with InfiniBand networking, aimed at teams running demanding workloads rather than single-GPU experiments.
GPUs and pricing
The lineup runs from A100 and L40S up to the latest Blackwell systems (HGX B200, GB200 NVL72, GB300 NVL72) alongside HGX H100 and H200. On-demand rates are quoted per 8-GPU node (per their pricing page, June 2026): an HGX H100 node lists at $49.24/hr, A100 at $21.60/hr, L40S at $18.00/hr, and HGX B200 at $68.80/hr. Spot instances cut roughly 35-50% off on-demand, and reserved capacity advertises up to 60% off for committed usage.
Networking and extras
Internal and internet data transfer are free, public IPs are $4/month, and Direct Connect ranges from $1,250 to $50,000/month by speed. The platform leans toward managed Kubernetes and Slurm for orchestration.
Who it fits
CoreWeave fits teams that need a lot of GPUs wired together with fast interconnect, for training large models or running high-throughput inference. It is less oriented toward someone who just wants a single GPU by the hour. For that, a marketplace like Vast.ai or a per-minute provider like Lambda or Hyperstack is a closer match.
CoreWeave Alternatives
Explore 115 products in the Inference APIs category. View all CoreWeave alternatives.
AiQu
Swedish GPU infrastructure and LLM hosting platform with API-first deployment, no Kubernetes required
Beam
Open-source serverless GPU cloud with sub-second cold starts and auto-scaling
SambaNova
Custom AI chip inference platform with purpose-built hardware for high-throughput LLM serving
Work on CoreWeave? Feature it at the top of Inference APIs.
Is your product missing?