CoreWeave Cloud review

CoreWeave Cloud is an AI-focused cloud platform for training and serving models at scale. It brings together GPU and CPU compute, storage, networking, and tools for managing AI infrastructure and workloads. The listed compute options include NVIDIA GPUs and AMD and Intel CPUs, with configurations ranging from single-GPU instances to multi-GPU systems.

For infrastructure management, CoreWeave Kubernetes Service is managed Kubernetes designed for AI workloads. The platform also includes AI-focused object storage and dedicated VAST clusters, plus high-bandwidth, low-latency networking for clusters. CoreWeave Forge adds tools for model training, fine-tuning, inference, and agent development.

The pricing page lists hourly on-demand and spot rates, not monthly or annual plans. That makes the buying model centered on choosing an instance and paying by the hour, rather than selecting a published subscription tier. Some instance rates are public; others require a sales quote.

The catalog covers a wide range of AI hardware, from single-GPU machines to eight-GPU systems and large-scale training configurations. The pricing page lists an NVIDIA GH200 with one 96 GB GPU at $6.50 per instance-hour on demand. For eight-GPU options, it lists the HGX H100 at $49.24/hour on demand and the A100 at $21.60/hour. Spot rates are lower on several configurations, including the H100 at $19.71/hour in North America and $19.51/hour in Europe.

There are also CPU instances, with the lowest listed rate at $3.27/hour for an Intel Emerald Rapids spot instance. That is a CPU rate, not an entry-level GPU price. The range gives teams options for different workloads, while the listed local storage varies by configuration from 7.68 TB to 61.44 TB.

The platform combines compute with storage, networking, managed Kubernetes, and Forge’s model tools. That can reduce the need to assemble those components from separate providers, though the published pricing information does not give separate rates for storage or networking.

Hourly pricing makes costs harder to compare with providers that publish monthly plans. CoreWeave’s pricing page does not list monthly or annual billing, and no free plan is listed; a free trial is not specified. The lowest published figure, $3.27/hour, is for an Intel CPU spot instance, not GPU compute. GPU prices vary considerably by model and billing type, so the relevant rate depends on the workload and selected hardware.

Some important rates are sales-gated. The page lists on-demand pricing for the GB300 NVL72, HGX B300, and standard-memory RTX PRO 6000 as “contact sales.” It also lists contact-sales pricing for some CPU instances. Spot availability is not uniform: for example, the L40 and L40S spot rates are shown for North America but not Europe, while several newer systems have no spot rate listed.

The listed instance specifications are concrete, but the published pricing information does not show separate charges for storage or networking. Teams should confirm those costs and any sales-quoted rates before comparing total spend.

CoreWeave Cloud has no free plan, and the listed rates are hourly—not monthly or annual. The lowest published rate is $3.27 per instance/hour for an Intel Emerald Rapids spot instance; its on-demand rate is $5.31. Spot pricing is not guaranteed to be available, and the pricing details don’t show a spot rate for that Intel configuration.

GPU costs vary by instance and rate type. An NVIDIA A100 is listed at $21.60/hour on-demand or $9.65/hour spot in North America. An HGX B200 jumps to $68.80/hour on-demand, with spot rates of $34.11 in North America and $34.87 in Europe. Some on-demand prices require contacting sales, including GB300 NVL72 and HGX B300.

Inference rates are listed separately for North America, including $2.70/hour for A100 and $10.50/hour for GB200. Other listed inference rates require contacting sales. Budget around the hourly rate and verify region, availability, and sales-gated prices before estimating spend.

Treat a move to CoreWeave as an infrastructure migration, not just a GPU swap. Match your current workloads to CoreWeave’s GPU or CPU instances, then account for storage, networking, managed Kubernetes, and any use of CoreWeave Forge. The published material doesn’t publish migration tooling, data-export terms, or egress charges, so confirm those details with CoreWeave before planning a cutover.

For a move away, identify which workloads depend on CoreWeave-specific services and make a plan for the data and model artifacts they use. The published information doesn’t establish how easy export is or which formats and tools are supported. It also includes no Reddit reports about why customers leave, so we can’t attribute common departure reasons to users. Before switching, compare the exact instance configuration and regional rates you need; spot prices and some on-demand rates vary by region or require a sales quote.