Home / Comparisons / CoreWeave
CoreWeave
Large-scale AI cloud, Kubernetes-native, enterprise contracts
What is CoreWeave?
CoreWeave is large-scale AI cloud, Kubernetes-native, enterprise contracts. Visit CoreWeave's site.
What's good about CoreWeave
- Fleet scale and hardware breadth, including GB200 NVL72 capacity most brokers cannot source
- Kubernetes-native infrastructure with high-performance storage and networking for large distributed training
- No ingress, egress or Kubernetes control-plane fees; reserved capacity up to 60% off
CoreWeave limitations
- HGX instances sell in blocks of 8 GPUs (GH200 is the one single-GPU exception, at $6.50/hr)
- No self-serve signup: access runs through a sales conversation
- Billing splits GPU, CPU, RAM and storage into separate line items
CoreWeave pricing
CoreWeave figures are On-demand, North America, per GPU, read from their own pricing page on 2026-09-19. The Aquanode column is the lowest live per-GPU rate in our marketplace feed and moves on its own. The two columns are not measured the same way, so treat this as a starting point, not a quote.
Aquanode column last updated: 2026-11-04 UTCRefreshes every 6 hours
| GPU | CoreWeave | Aquanode (live) |
|---|---|---|
| GB200 NVL72 | $10.50/GPU/hr | – |
| HGX B200 | $8.60/GPU/hr | from $9.35/GPU/hr |
| HGX H200 | $6.31/GPU/hr | from $3.98/GPU/hr |
| HGX H100 | $6.16/GPU/hr | from $2.78/GPU/hr |
| A100 80GB | $2.70/GPU/hr | from $1.40/GPU/hr |
Source: https://www.coreweave.com/pricing, On-demand, North America, per GPU. Verified 2026-09-19. Vendors change prices; check theirs before deciding.
CoreWeave billing
- Billing model: Per-hour, on-demand or reserved; GPU, CPU, RAM and storage billed as separate line items
- Minimum commitment: 8 GPUs per HGX node (1 on GH200); its best rates need multi-year reserved contracts
Egress: No charge for ingress, egress or Kubernetes control-plane traffic.
Source: https://www.coreweave.com/pricing, read 2026-09-17.
CoreWeave vs Aquanode
CoreWeave is built for organisations buying GPU capacity at fleet scale on contracts, with Kubernetes-native infrastructure and one of the largest H100/H200/GB200 estates anywhere. If you are one developer who wants a box that remembers your environment, it is the wrong shape and the published per-GPU rates show it.
Where CoreWeave wins
- Fleet scale and hardware breadth, including GB200 NVL72: capacity most brokers simply cannot source.
- Kubernetes-native infrastructure, high-performance storage and networking built for large distributed training.
- Enterprise contracting, compliance and support that a self-serve product does not attempt.
Where Aquanode wins
- Your setup outlives the GPU you rented it on. Save the environment (custom nodes, model weights, packages, the config you spent an evening getting right) and bring it back up later on a completely different provider, instead of reinstalling it from scratch every session.
- Supply moved, the price moved, or the region ran dry. Your environment follows you out. It is not stranded in the account that happened to create it, which is the part that makes leaving any single vendor cheap.
- Pause and resume in place, the same mechanism everywhere: a snapshot-and-terminate, then a fresh box restored from that snapshot. The same environment also comes back by restoring your saved setup onto a fresh box, a little slower, same result.
- Turn on automatic snapshots yourself and pick the interval (as often as every 15 minutes, 30 by default) instead of remembering to snapshot by hand or wiring up your own cron job.
- One account, one bill and one set of keys across every provider we support, rather than a separate login and invoice per vendor every time you chase capacity.
- Run a saved version of your setup as a job. If a provider takes the box back mid-run, we detect the loss and queue the run to retry on a different provider, excluding the one that just lost it, and if the job checkpoints the next attempt picks up from the last one. Jobs scale to zero when the queue empties, so an idle one is not sitting on a rented GPU, and every job is bounded in the unit we bill: its time limit times its attempts times its machines is the most a single run can cost.
What we do not claim
Restoring an environment requires a snapshot that already exists. Stopping a deployment yourself captures it on the way out, so you can bring it back later on any provider. A provider-side termination is different: it is only recoverable if you had already switched automated snapshots on for that deployment, and it costs you the work since the last one. Automated snapshots are opt-in, nothing runs until you start it, and with none running there is nothing to restore.
`aq deploy --snapshot <id>` rents the cheapest matching GPU and restores your snapshot onto it, including onto a different provider than the one it came from. Creating the snapshot is a separate step today: the standalone ogre CLI on the box writes it.
CoreWeave: common questions
Should I use CoreWeave or Aquanode?
If you are contracting for reserved fleet capacity or running large distributed training, CoreWeave. If you are renting single boxes and the expensive part is rebuilding your environment each time, that is what we are for.
Does CoreWeave require a commitment?
Not for burst capacity. CoreWeave's pricing page lists an on-demand price per hour and a spot price per hour for its GPU configurations. Commitment is the discount lever rather than the entry ticket: CoreWeave states it offers up to 60% discounts over its on-demand prices for committed usage.
Source: https://www.coreweave.com/pricing, read 2026-09-02.
CoreWeave alternatives
Other GPU cloud provider reviews
Related reading: H100 pricing and specs, H100 vs H200, A100 vs H100, H100 and H200: SXM vs NVL vs PCIe, and The best GPUs for AI, ranked.
