Catalog
We list what we can as on-demand. A lot of CoreWeave’s interesting capacity is reserved. Those rows stay labeled.
Provider review / verified 2026-09-05
CoreWeaveGPU review
CoreWeave is a specialist GPU hyperscaler-in-all-but-logo. Kubernetes, InfiniBand, reserved clusters, and the GPUs that labs brag about. On-demand exists in pockets. The real product is capacity you plan.
Models listed6
From /GPU/hr$1.79
Regions in sample1
Business modelSpecialist GPU cloud
Our reading
Who this provider is for
I do not send a weekend LoRA to CoreWeave. I send teams who already know they will spend six figures and want NVIDIA in a real datacenter fabric. If your buying process is a credit card and a Discord screenshot, you will bounce. If your buying process is ‘we need N SXM nodes with IB and a date,’ this is one of the three doors I mention with a straight face. The other two are a hyperscaler and whoever is actually holding Blackwell this quarter.
Rent it if Scale-ups and enterprises training or serving at cluster size, who can wait for reserved capacity and talk to a human.
Skip it if Anyone optimizing a $20 experiment. Anyone who needs a GPU in four minutes with no conversation.
How we compare hoursStrengths
- Purpose-built GPU datacenters, not a side-hustle on leftover CPU estates
- InfiniBand and dense NVIDIA SKUs when you are actually clustering
- Kubernetes-native story that teams already running GPU operators will recognize
- The inventory people mean when they say ‘neocloud’ without meaning a marketplace
Limits
- Self-serve on-demand is not the heart of the business. Expect reserved and waitlists on the hot SKUs
- Pricing is not a Vast.ai leaderboard. Quotes happen. We still refuse to mix them with published hours
- You will feel the enterprise-ness: onboarding, networking, identity
- Overkill, and often unavailable, for single-GPU fine-tunes
Same standard, every cloud
What we measured on CoreWeave
Four checks we apply to every provider. This is not a score, and it is not a paid ranking.
Price signal
Published hours sit above marketplace floors. The comparison that matters is against AWS/GCP reserved and other neoclouds, not a 4090.
Product shape
Kubernetes, IB fabrics, GPU nodes. This is infrastructure for people who have already chosen Kubernetes.
Still verify
Ask what is actually reservable this month. A blog post about GB200 is not a node you can SSH to.
- Model
- Specialist GPU cloud
- Headquarters
- Roseland, US
- Setup
- Kubernetes, VMs, reserved clusters
- Interconnect
- InfiniBand / NVIDIA fabrics
- On-demand
- Limited vs reserved
- Self-serve speed
- Slower than pods
- Founded
- 2017
- Best known for
- AI supercloud capacity
Catalog listings
CoreWeave catalog sample
Published on-demand hours first. Spot and waitlists are labeled. Confirm the live rate before you provision.
GPUsRegionBilling typeInterconnect/GPU/hr
This is not a pod host
If you arrived from RunPod, reset your expectations. CoreWeave is closer to ‘AWS but GPU-first’ than to a template marketplace. That is why the website talks about clients you have heard of. That is why you may not get a GPU tonight.
I respect that. I also will not pretend it is the same SKU as a Community H100.
Reserved is the honest product
The industry’s worst habit is publishing a reserved number next to a marketplace on-demand number. On GPUBeacon, CoreWeave reserved stays reserved. If we show an on-demand hour, it is because it was published as one.
If you need a fleet next quarter, talk to them. If you need a GPU before lunch, talk to RunPod or Packet.ai.
When I pick CoreWeave over Lambda or AWS
Over Lambda: when the cluster is the product and IB topology is not a footnote. Over AWS: when I want GPU-native ops and I am willing to give up the rest of the AWS religion, IAM spaghetti, credits, every service in the slide.
If the rest of the company already lives in AWS, staying there can be cheaper in human time even when the GPU hour loses. I say that as someone who hates saying it.
How I actually rent it
- 01
Know your topology
GPU count, IB, storage, region. Do not lead with a GPU nickname.
- 02
Ask reserved vs on-demand
Get the commercial shape in writing. We will not disguise a quote as a catalog hour.
- 03
Compare two neoclouds and one hyperscaler
CoreWeave, Lambda or Crusoe, and the cloud you already have an org in.
FAQ
CoreWeave questions
Is CoreWeave cheaper than AWS?+
Sometimes on raw GPU hours, especially reserved. Rarely when you count the engineering you already did on AWS. Compare GPU-only and total cost separately.
Can I get an on-demand H100 on CoreWeave today?+
Treat it as maybe. Hot SKUs waitlist. Confirm live stock. Do not plan a demo on a press release.
CoreWeave vs Lambda?+
CoreWeave for cluster-scale reserved AI infrastructure. Lambda for self-serve ML cloud with workstation culture. Different doors.
Is CoreWeave good for inference?+
Yes at scale, on their terms. For a hobby endpoint, it is a cannon. Use RunPod serverless or Together.
Who should skip CoreWeave?+
Anyone still choosing between a 4090 and an A6000. Come back when the job has a cluster diagram.
Sources
Where these numbers come from
Prices and SKUs are from the GPUBeacon catalog sample. Ranking rules are on the methodology page. Outbound links may be affiliate. They never change the hour we display.
GPUs
Keep comparing
Other providers to compare
Same fields, different product. Compare the GPU hour before you lock a brand.
