CoreWeave review
Updated September 28, 2026 · DeployCue Editorial Team · Facts checked September 26, 2026
We earn a commission when you sign up through links on this site. Partner status can affect where a brand appears in our rankings. Partners with a live offer in your country come first, then every other brand by score.
Our verdict
CoreWeave is a specialist GPU cloud rather than a cheap general one. It suits teams that need current NVIDIA hardware and free outbound data, and it sits at 93 of 95 on entry price, so the cheapest hour is not the selling point.
Key facts
- Best for
- Teams training on current NVIDIA datacenter hardware with heavy data movement.
- GPU models
- NVIDIA GB300 NVL72, NVIDIA GB200 NVL72, NVIDIA HGX B300, NVIDIA HGX B200, NVIDIA RTX PRO 6000 Blackwell Server Edition (High Memory), NVIDIA RTX PRO 6000 Blackwell Server Edition (Standard Memory), NVIDIA GH200, NVIDIA HGX H100, NVIDIA HGX H200, NVIDIA L40, NVIDIA L40S, NVIDIA A100, Tesla V100 NVLINK, RTX A5000, RTX A6000, A40, A100 40GB PCIe, A100 40GB NVLINK↗
- GPU hour from
- $6.50↗
- Egress fee
- Free data transfer to the internet and within CoreWeave; egress and IOPS free↗
- Locations
- from 51↗
- Interconnect
- Direct Connect (dedicated 10G/100G/400G and virtual DX), NVIDIA Quantum InfiniBand up to 3,200 Gbps per node, high-speed fiber backbone↗
- Legal entity
- CoreWeave, Inc.↗
- Headquarters
- United States↗
- Founded
- 2017↗
- Support channels
- Email, Help-centre↗
- Pricing model
- Hourly billing
- Official website
- coreweave.com
Apps and profiles
Pricing
CoreWeave NVIDIA RTX PRO 6000 Blackwell (USD)
compute: $2.50 / hour
CoreWeave NVIDIA RTX PRO 6000 Blackwell (USD)
compute: $1.20 / hour
CoreWeave NVIDIA L40S (USD)
compute: $2.25 / hour
CoreWeave NVIDIA L40S (USD)
compute: $0.99 / hour
CoreWeave NVIDIA L40 (USD)
compute: $1.25 / hour
CoreWeave NVIDIA L40 (USD)
compute: $0.78 / hour
CoreWeave NVIDIA H200 (USD)
compute: $6.30 / hour
CoreWeave NVIDIA H200 (USD)
compute: $2.58 / hour
CoreWeave NVIDIA H100 (USD)
compute: $6.16 / hour
CoreWeave NVIDIA H100 (USD)
compute: $2.44 / hour
CoreWeave NVIDIA GH200 Grace Hopper (USD)
compute: $6.50 / hour
CoreWeave NVIDIA GB200 (USD)
compute: $10.50 / hour
CoreWeave NVIDIA B300 (USD)
compute: $4.48 / hour
CoreWeave NVIDIA B200 (USD)
compute: $8.60 / hour
CoreWeave NVIDIA B200 (USD)
compute: $4.26 / hour
CoreWeave NVIDIA A100 80GB (USD)
compute: $2.70 / hour
CoreWeave NVIDIA A100 80GB (USD)
compute: $1.19 / hour
Pros and cons
Pros
- Data transfer to the internet and within CoreWeave is free, and IOPS are free as well.
- The control plane is free, so Kubernetes management adds no separate fee.
- The range in NVIDIA GB300 NVL72, NVIDIA GB200 NVL72 and NVIDIA HGX B300 spans rack-scale Blackwell systems, HGX cards and the RTX PRO 6000 Blackwell Server Edition.
- At least from 51 AI data centres give a wide choice of placement.
Cons
- GPU hours at $6.50 make short jobs expensive, and CoreWeave places 93 of 95 on that measure.
- Support runs on email and help centre only, with no phone line for urgent outages.
- Billing is limited to the hourly model, so short tests pay for a full hour rather than metered seconds.
Who it suits
- Teams training or serving frontier models that need current NVIDIA datacenter hardware.
- Organisations moving large datasets in and checkpoints out, since egress carries no charge.
- Kubernetes teams that want a free control plane with node capacity priced at $3.36.
Who should look elsewhere
- Budget-first GPU buyers, who should weigh ComputePrices.com, where H100 SXM, A100 SXM and H200 covers a wider mix including AMD and consumer cards.
- Salad, whose NVIDIA RTX 3060, NVIDIA RTX 3070 and NVIDIA RTX 3070 Ti line-up centres on consumer and community cards rather than datacenter racks.
- SberCloud, which lists B300, H100 and A100 against the range in CoreWeave's NVIDIA GB300 NVL72, NVIDIA GB200 NVL72 and NVIDIA HGX B300 and a smaller footprint than from 51 AI data centres.
Core features
CoreWeave rents NVIDIA accelerators from NVIDIA GB300 NVL72, NVIDIA GB200 NVL72 and NVIDIA HGX B300, spanning rack-scale Blackwell systems down to HGX, L40S and A100 cards. Capacity sits in at least from 51 AI data centres, and the fabric is built for training clusters: Direct Connect with dedicated and virtual links, NVIDIA Quantum InfiniBand, and a high-speed fiber backbone.
The company behind it
CoreWeave, Inc. is a US company founded in 2017 and built around AI data centres rather than general-purpose cloud hosting. Day to day help comes through email and help centre, so answers arrive by written channel or self-serve documentation rather than by phone. Teams needing live voice support should check those channels before committing.
How CoreWeave compares
CoreWeave vs RunPod8.9 vs 9.5
CoreWeave vs ScalaHosting8.9 vs 8.3
CoreWeave vs Contabo8.9 vs 9.2
CoreWeave vs Kamatera8.9 vs 9.5
CoreWeave vs Cloudways8.9 vs 9.2
CoreWeave vs PlanetHoster8.9 vs 8.7
CoreWeave vs Acronis8.9 vs 9.9
CoreWeave vs Aruba8.9 vs 9.3
CoreWeave vs Decodo8.9 vs 9.1
CoreWeave vs Hostinger8.9 vs 9.1- CoreWeave vs MarsProxies8.9 vs 9.0
CoreWeave vs hide.me8.9 vs 8.7
CoreWeave vs Proton Drive8.9 vs 8.3
Where CoreWeave is available
- Available
- Everywhere except Belarus, Cuba, Iran, North Korea, Russia, Syria
Countries from the brand (Source, checked September 26, 2026)
Frequently asked questions
What does an hour of GPU capacity cost?+
CoreWeave's entry GPU price is $6.50, and that sits above the €0.63 median across GPU providers on this site. Its rank on entry price is 93 of 95, so treat the cheapest card as the floor rather than a typical training cost.
Is there a data transfer charge?+
CoreWeave does not bill for data transfer to the internet or for traffic inside its own network, and IOPS are free as well. The site median egress price is €0, so free outbound data is worth real money for jobs that move datasets in and checkpoints out.
Does the Kubernetes control plane cost anything?+
The control plane is free, matching the 66.7% share of providers here, while the site median control plane price is €0.06. Node capacity is separate and starts at $3.36, so a cluster's real cost sits in the worker bill.
How is capacity billed?+
Capacity is billed on the hourly model, which is the only one listed and matches 38.5% of providers on this site. Per-second metering does not appear in that hourly model, so short benchmarks and tests pay for a whole hour.
Which GPUs can you rent?+
The catalogue runs across NVIDIA GB300 NVL72, NVIDIA GB200 NVL72 and NVIDIA HGX B300, so rack-scale Blackwell systems, HGX-class cards and the A100, H200 and L40S all sit alongside each other. H100 SXM, A100 SXM and H200 offers a wider mix that also includes AMD and consumer cards.
Where does the hardware run?+
CoreWeave, Inc. is headquartered in the United States, founded in 2017, and runs at least from 51 AI data centres, well above the 6 median across GPU providers. Teams needing placement choices or local latency get the most from that spread.
How do you reach support?+
Support runs through email and help centre, so help arrives by written channel or self-serve documentation rather than by phone. The 90.2% share of providers here lean on a help centre, and the phone channel that only 43.6% of them offer is not among CoreWeave's routes.
Sources
- docs.coreweave.com/llms.txt checked September 26, 2026
Backs: Support channels
“Contact support: Contact CoreWeave support through the Cloud Console help portal or by email for technical assistance”
- docs.coreweave.com/policies/terms-of-service/security-and... checked September 26, 2026
Backs: Legal entity
“How CoreWeave, Inc. keeps its infrastructure and customer service data secure”
- en.wikipedia.org/wiki/CoreWeave checked September 26, 2026
Backs: Founded
“Founded September 21, 2017 (as Atlantic Crypto) in New Jersey”
- coreweave.com/ai-data-centers checked September 26, 2026
Backs: Locations
“CoreWeave operates in 51 data centers across North America and Europe”
- coreweave.com/pricing checked September 26, 2026
Backs: Egress fee, Free control plane, GPU hour from, GPU models, Headquarters, Interconnect, Pricing model, Node price from
“Data transfer between CoreWeave and internet Free; Data transfer within CoreWeave Free; Egress and Input Output Operations per Second Free”
Compiled from 5 source pages, checked September 26, 2026. We did not open an account.
Not confirmed yet: Billing granularity.
Related by the facts
CoreWeave
Teams training on current NVIDIA datacenter hardware with heavy data movement.