Saaskly

CoreWeave

CoreWeave is a Nasdaq-listed AI cloud renting whole eight-GPU NVIDIA nodes with InfiniBand, Kubernetes and Slurm to teams training and serving large models.

GPU & High Performance ComputeHow we test
2.9
Saaskly score / 5
10th of 12 in GPU & High Performance Compute

Who owns it

CoreWeave, Inc. is a US company listed on Nasdaq as CRWV, with its headquarters in Livingston, New Jersey. It was founded in 2017 as Atlantic Crypto by Michael Intrator, now chairman and chief executive, Brian Venturo and Brannin McBee, and renamed CoreWeave in 2019 when it moved from Ethereum mining to renting out GPUs. It floated on 27 March 2025 at $40.00 a share, below the expected range, and closed its first day flat. NVIDIA held roughly 7% before the listing and invested a further $2bn at $87.20 a share in January 2026, taking its holding to around 11%. The FY2025 accounts show revenue of $5.1bn with Microsoft at approximately 67% of it and 68% of year-end receivables, alongside $60.7bn of remaining performance obligations, so this is a company whose finances lean heavily on a handful of very large customers. Its $9bn all-stock bid for Core Scientific was announced on 7 July 2025 and rejected by Core Scientific stockholders on 30 October 2025; the two remain commercial partners. The acquisition of Weights and Biases did complete, on 5 May 2025. A UK buyer can contract with CoreWeave UK Limited, company number 15272646, incorporated on 9 November 2023 and registered at 133 Park Street, London SE1 9EA. CoreWeave owns and runs its own GPUs, networking and storage, but mostly leases the buildings: Crawley sits with Digital Realty, London Docklands with Global Switch, and US capacity comes through long leases with Applied Digital, Flexential and others. Its security page says it requires colocation providers to meet SOC 2 and ISO 27001 frameworks "throughout their contract lifecycle".

## What it costs

Prices are in US dollars for both the North American and European price lists, with no GBP option and no VAT position stated, read on 22 September 2026. The important structural fact is that CoreWeave sells nodes, not GPUs. An HGX H100 instance is eight H100s with 128 vCPU, 2,048 GB of RAM and 61.44 TB of local NVMe on an InfiniBand fabric, and it costs $49.24 an hour, which works out at about $6.16 per GPU-hour and $35,945 a month at 730 hours. There is no way to buy one H100. The same shape applies across the range: H200 at $50.44 an hour, B200 at $68.80, A100 at $21.60, L40S at $18.00, and the cheapest node of any kind, the L40 at $10.00 an hour or $7,300 a month, which is $1.25 per GPU-hour. GB200 NVL72 is $42.00 an hour for a four-GPU instance and GB300 is contact sales only. CPU-only nodes start around $6.42 an hour for an AMD Genoa machine. Spot is properly cheap and, unusually, published: $19.71 an hour for the H100 node in North America and $19.51 in Europe, $9.65 and $9.51 for A100, which is about 60% off on demand. Committed pricing is not published. The only public figure is "up to 60% discounts over our On-Demand prices for committed usage", with no term lengths, no minimum quantity and no minimum spend, and CoreWeave's documentation says Reserved Instances and Flex Reservations "are set up commercially through your Account Executive". Reserved capacity bills flat whether you use it or not; Flex Reservations charge a holding rate plus a usage rate.

Storage is $0.070 per GB a month for distributed file, and object storage runs at $0.06 hot, $0.03 warm and $0.015 cold per GB a month. Block storage is offered but not priced publicly. Data transfer costs nothing, to the internet or between CoreWeave regions, which is the single most useful pricing fact here because it removes the per-gigabyte egress bill that catches people out elsewhere. Prices are published per hour and CoreWeave does not state a finer billing increment, nor whether payment is by card or invoice, nor whether a stopped on-demand instance is billed. For the sort of buyer this site is written for, there is no small monthly bill to calculate: the floor is an eight-GPU node, so a month on the cheapest L40 node is $7,300 and a month on an H100 node is $35,945, before storage.

## What you get and what to watch

A new account cannot launch anything today. CoreWeave's documentation states that "When an organization is first approved to join CoreWeave, the CoreWeave Sales team sends an invitation email to the creator of the organization to activate their account", so the organisation is created by CoreWeave rather than by the customer. The pricing page's buttons are "Contact us", "Contact sales" and "Talk to an expert", with no signup. Once the account exists, an administrator creates an API token, downloads a kubeconfig and installs kubectl and Helm before deploying, and from there Spot Node Pools are self-serve while Reserved and Flex capacity go back through the account executive. CoreWeave does not publish how long any of that takes, saying only that "Your Account Executive gives you the authoritative timeline for a specific request". Its ARENA programme offers evaluation access to production hardware, but it is a governed trial arranged with CoreWeave, not a free tier. There is no published free credit.

The UK footprint is genuine. Crawley, with Digital Realty, went live in October 2024 running NVIDIA H200 GPUs on Quantum-2 InfiniBand, and London Docklands, with Global Switch, followed in December 2024, both on renewable power, with London named as the European headquarters. CoreWeave has committed roughly £2.5bn to the UK in stages, starting at £1bn, rising by £750m in October 2024 and by a further £1.5bn in September 2025. On the continent, a $2.2bn commitment covers three data centres in Norway, Sweden and Spain, and their operational status is not confirmed on the site. The pricing page does sell a European region with its own spot rates. On the platform you get bare-metal nodes, CoreWeave Kubernetes Service, SUNK for running Slurm on Kubernetes, a non-blocking fat-tree InfiniBand fabric with SHARP, object and distributed file storage, an official Terraform provider covering CKS clusters and VPCs, the cwic command line tool and an API. A documentation MCP endpoint exists for querying the docs, but no first-party MCP server for provisioning infrastructure was found. On compliance, the Trust Center lists ISO 27001, SOC 2 Type II, ISO 27017 and ISO 27018, but the reports themselves are available only under NDA through an account team, and the documentation compliance page sits behind an access code. The CLOUD Act is not addressed anywhere public, and the parent is a US-listed corporation whatever entity you contract with.

Support runs through a named account executive and a support portal rather than a public queue, and CoreWeave does not publish support tiers, hours or response times. The terms of service do publish uptime targets: 99.9% or better for instances across multiple regions and 99% or better for a single region, with credits of 10%, 25% or 50% of the monthly charge depending on how far below target the month falls, capped at 50%. One oddity is worth knowing: no financial credits have been available for the EWR1 data centre since 1 August 2022. Object storage carries its own 99.9% agreement, which applies only to the cwobject.com endpoint. There is no consumer review signal to report: CoreWeave's Trustpilot profile is unclaimed with zero reviews, and G2 blocked automated access. No regulatory action, breach or data-loss incident was found, though litigation around the flotation and the failed Core Scientific deal was not searched.

Not yet tested by Saaskly. Assessed on published pricing, vendor documentation, company filings, the vendor's own status and trust pages, and third-party reviews.

  • +Two live UK data centres, Crawley with Digital Realty since October 2024 and London Docklands with Global Switch since December 2024, plus a UK contracting entity, CoreWeave UK Limited, number 15272646
  • +Spot capacity is published and self-serve once the account exists: $19.71 an hour for an eight-GPU H100 node against $49.24 on demand, about 60% off
  • +Data transfer is free to the internet and between CoreWeave regions, and the terms publish a 99.9% uptime target for multi-region instances with credits of 10% to 50% of the monthly charge

Cons

  • −The smallest H100 purchase is a whole eight-GPU node at $49.24 an hour, about $35,945 a month at 730 hours, so a team that needs one GPU pays for eight
  • −There is no self-serve signup: CoreWeave's documentation says the sales team approves the organisation and sends the activation invitation before anyone can log in
  • −Committed pricing is not published beyond "up to 60%", with no term lengths, no minimum quantity and no minimum spend, so the entry point cannot be sized without a sales call
  • −Certification reports are available only under NDA through an account team, and the documentation compliance page returned an access code wall on 22 September 2026

Pricing

Minimum commitment: $35,945 ($35,945/mo × 1 seat × 1 month (no commitment)). The least you can spend before walking away; excludes numbers, minutes and add-ons.

HGX H100 80 GB, eight-GPU node

$35,945/mo · no commitment

  • ✓Eight H100 80 GB with 128 vCPU, 2,048 GB RAM and 61.44 TB local NVMe on InfiniBand, at $49.24 an hour for the whole node, about $6.16 per GPU-hour and $35,945 a month at 730 hours. Single GPUs are not sold. Spot is $19.71 an hour in North America and $19.51 in Europe. Committed pricing is "up to 60%" off but the rate and term are not published. USD, VAT not stated, read 22 Sep 2026.

L40, eight-GPU node

$7,300/mo · no commitment

  • ✓Eight L40 with 128 vCPU, 1,024 GB RAM and 7.68 TB local NVMe at $10.00 an hour, $7,300 a month at 730 hours, which is $1.25 per GPU-hour and the cheapest way into CoreWeave. Distributed file storage $0.070 per GB a month, object storage from $0.015; data transfer free.

Specifications

Rating
2.9
Pricing and plans
Committed H100Sales only
Cheapest dedicated server$4,687/mo (AMD Genoa 9274F CPU-only node)
Setup feeNone
BillingPer hour
EgressIncluded
Committed pricingSales only
GPU dedicated server$7,300/mo (Eight L40, 128 vCPU, 1,024 GB RAM, 7.68 TB local NVMe)
Free creditNone found. ARENA is an evaluation programme arranged with CoreWeave, not a free tier.
Features
H100 per GPU-hour*$6.16/h (NVIDIA H100 80 GB SXM (sold only as an eight-GPU HGX node))
Cheapest GPU$1.25/h (NVIDIA L40 (sold only as an eight-GPU node))
8xH100 node$49.24/h
Spot / interruptible✓
Storage (per GB/mo)$0.070
Self-serve signupNo
GPU quotaQuota ticket
UK location✓
H100 in the UKNo
EU locations0
Locations7
Cluster networking✓
Kubernetes✓
Slurm✓
Bare metal✓
Dedicated server termNone for on demand; Reserved and Flex terms not published
Stopped instance billed✓
Sales-only aboveEverything (sales-led)
UK citiesCrawley, London
NVIDIA GPUs✓
AMD GPUsNo
Multi-GPU nodes✓
Dedicated serversNo
Object storage✓
Block storage✓
Private networking✓
Integrations and developers
API✓
Terraform✓
MCP serverNo
CLI✓
Ownership and trust
Hardware**Own hardware, colocated
Owned inUS-owned
OwnerIndependent
Uptime SLA (%)99.90
HQUS
Publicly listed✓
Sovereign offeringLocal contracting entity
ISO 27001✓
SOC 2✓
2FANo
Support
Phone supportNo
Support channelsNamed account executive, Support portal

* One NVIDIA H100 80 GB per GPU-hour, on demand, in the currency the vendor shows a UK visitor; where the vendor sells no H100 the nearest GPU is named. Committed prices are the cheapest published rate and its minimum term.

** Who owns and runs the GPUs. "Third-party hosts" means the vendor is a marketplace for other people's machines and the SLA is theirs, not the vendor's.

Researched and drafted with AI assistance; reviewed and approved by Tim Meredith. How we use AI

Work at CoreWeave? Something not look right? Register in the vendor portal to see what we publish and send corrections.