CoreWeave Pricing for Startups and Developers: 2026 Breakdown
CoreWeave pricing is straightforward: pay per hour for GPU instances, per GB-month for storage, and per GB for egress. No reserved instance commitments required for on-demand, no complex pricing tiers. For startups and developers doing AI work, CoreWeave is typically 40-70% cheaper than equivalent AWS or GCP GPU instances.
💰 Free Trial: CoreWeave offers $50 in free credits for new developer accounts — enough to run several training experiments on A100 instances before spending a dollar. Sign up at coreweave.com.
How CoreWeave Pricing Works
CoreWeave bills by the second for compute, and there are no minimum commitments for on-demand instances. You provision a GPU pod, it starts billing when the pod is Running, and stops when you delete it. This is fundamentally different from AWS reserved instances or committed use discounts — you only pay for actual compute time.
Key pricing categories:
- GPU Compute: Per-hour rate for each GPU type — your biggest cost item
- CPU Pods: For non-GPU workloads like data preprocessing — much cheaper than AWS EC2
- Storage (PVC): Per GB-month on HDD or SSD tiers
- Network Egress: Data out of CoreWeave — usually small for training workloads
Full GPU Pricing Table
| GPU | VRAM | CoreWeave/hr | AWS On-Demand/hr | Saving |
|---|---|---|---|---|
| H100 SXM5 80GB | 80GB | $2.06 | $3.40 (p5) | 39% |
| H100 NVL 94GB | 94GB | $3.54 | N/A comparable | — |
| A100 80GB SXM4 | 80GB | $2.21 | $4.10 (p4de) | 46% |
| A100 40GB PCIe | 40GB | $2.06 | $3.91 (p4d) | 47% |
| RTX A6000 | 48GB | $1.28 | ~$2.50 (g5) | ~49% |
| L40S | 48GB | $1.33 | ~$2.80 | ~52% |
Storage and Network Pricing
- NVMe SSD (fast, GPU-local): ~$0.07/GB-month — ideal for training datasets
- HDD (bulk storage): ~$0.04/GB-month — good for model archives
- Shared filesystem (NFS): ~$0.12/GB-month — multi-pod access during training
- Network egress: First 1TB free/month, then ~$0.05/GB — typically minimal for training
How CoreWeave Costs Compare
A real example: fine-tuning Llama 3.1 8B for 24 hours on 4x A100 80GB:
- CoreWeave: 4 x $2.21 x 24 = $212
- AWS p4d.24xl (8x A100): $32.77 x 24 = $786 (and you only need 4 GPUs)
- Google Cloud A100: ~$3.67 x 4 x 24 = $352
CoreWeave is the most cost-efficient option for AI training in 2026. For context on why this matters for the AI industry, see why CoreWeave is growing so fast.
CoreWeave vs AWS for AI Workloads 2026
Beyond pricing — GPU availability, performance, and the complete head-to-head comparison for production AI workloads.
Read Full Comparison →FAQs
CoreWeave offers $50 in free trial credits for developer accounts — enough for several hours of A100 or A6000 time. There is no permanent free tier like AWS Free Tier. After the trial, it's purely pay-as-you-go. For students and researchers, CoreWeave has academic program partnerships worth checking on their website.
Yes — CoreWeave offers Reserved Instance pricing for 1-year and 3-year commitments, typically 30-50% below on-demand rates. Enterprise agreements include custom pricing for large clusters. For startups burning through significant GPU time monthly, the reserved pricing often makes sense once your workloads are predictable.