The AI Stack
Sign in

CoreWeave

AI hyperscaler providing GPU cloud compute for large-scale AI training and inference.

Updated April 2026

Overview

Founded
2017
Headquarters
Livingston, NJ
Segment
GPU Cloud / AI Compute

Product overview

CoreWeave offers GPU compute, high-performance storage, networking, managed Kubernetes/Slurm services, and platform tools optimized for AI workloads on bare-metal NVIDIA GPUs. Key customers include OpenAI, Microsoft Azure, Meta, Mistral AI, IBM, and Jane Street. It differentiates via hypervisor-free architecture for superior performance, advanced cluster health automation (96% goodput), 24/7 expert support, and early NVIDIA GPU access, outperforming hyperscalers in AI reliability and speed.

Revenue model

À la carte on-demand GPU pricing (e.g., A100 80GB $2.21/hr GPU-only, H100 PCIe $4.25/hr, 8x H100 HGX ~$49/hr total; full instances $3+/hr) plus CPU/RAM/storage ($0.015-$0.06/GB/mo); up to 60% discounts for commitments; primary revenue from 4-6 year fixed-price enterprise contracts with prepayments.

Moat

CoreWeave's competitive moat is built on specialized bare-metal GPU infrastructure optimized for frontier AI model training, combined with long-term committed contracts (96% of 2024 revenue) that create predictable cash flows for capital-intensive expansion. This is reinforced by early access to cutting-edge NVIDIA hardware, proprietary software integration (Mission Control platform and Weights & Biases acquisition), and customer lock-in through switching costs—as technically sophisticated frontier labs become dependent on CoreWeave's homogeneous GPU clusters and integrated AI lifecycle management tools.