CoreWeave vs Thunder Compute (September 2025)
1. Pricing snapshot – A100 80 GB on‑demand
Thunder Compute rate is their current published price for on-demand A100 80 GB. CoreWeave pricing is based on their public pricing table and aggregator listings for single-GPU SXM SKUs.
2. Billing & minimum commitment
- Thunder Compute bills per second with no minimum; ideal for quick experiments or CI jobs.
- CoreWeave advertises on-demand hourly pricing; usage is metered hourly, so short sessions pay for the full hour.
Takeaway: If you spin up notebooks for <60 minutes at a time, Thunder Compute’s fine-grained meter typically saves an extra 5–25 %.
3. Developer experience
Thunder Compute focuses on bottoms-up engineers who want to “click → code.” CoreWeave targets ops teams comfortable managing clusters.
4. Storage & data management
Thunder Compute’s persistent disk is automatically attached to every instance, making snapshots and restarts seamless. CoreWeave’s storage is cheaper, but may require more manual setup and management.
5. Lock‑in & contracts
- Thunder Compute has no term commitments and supports standard SSH plus a convenient VS Code extension
- CoreWeave offers marketplace reservations and capacity contracts (commonly 1–3 years for volume discounts) – useful for large clusters, but adds vendor lock-in.
6. When Thunder Compute wins
- You need cheap A100s for interactive fine‑tuning or prototyping
- Jobs are bursty or short-lived – per-second billing avoids wasted spend
- Your team prefers VS Code over YAML manifests
- You want to live-resize RAM, vCPUs, and storage without redeploying
7. When CoreWeave wins
- You already run Kubernetes at scale and want full cluster control
- You need InfiniBand multi-GPU nodes or H100 clusters today (Thunder’s H100 launch is slated soon at $1.47/hr)
- Storage costs dominate and you have workflows built around cheap block and object storage tiers
Bottom line
At $0.78 per A100 80 GB, Thunder Compute undercuts CoreWeave by roughly 3× on raw GPU price and adds second-by-second billing plus VS Code convenience. CoreWeave still shines for teams that need large, reserved clusters and are happy living in Kubernetes. For most startups and indie ML engineers, Thunder Compute is the faster, cheaper on-ramp.