CoreWeave

Short Intro: CoreWeave is a public GPU cloud company (Nasdaq: CRWV) founded in 2017 as Atlantic Crypto Corp. by four energy futures traders: Michael Intrator, Brian Venturo, Brannin McBee, and Peter Salanki. The company began with GPU mining on a pool table in a hedge fund office, acquired distressed GPU inventory during the 2018 crypto winter, and pivoted to AI infrastructure when researchers started posting on forums about six-month waits for GPU time. CoreWeave generated $5.13 billion in revenue in 2025 and operates 250,000+ NVIDIA GPUs across 32+ data centers, with Microsoft, OpenAI, Meta, and Anthropic among the nine of ten top AI model providers that use the platform.

Key Capabilities:

• Bare-metal GPU access with no virtualization overhead across NVIDIA Blackwell, Hopper, and Ada architectures

• First cloud provider to bring up NVIDIA Vera Rubin NVL72 (June 2026)

• HGX B300 generally available with ultra-high density InfiniBand cluster configurations

• CoreWeave Kubernetes Service for autoscaling at node and workload layers

• SUNK unified orchestration combining Kubernetes and Slurm for training workloads

• SUNK Anywhere and LOTA Cross-Cloud for multi-cloud AI training across AWS, GCP, and Azure

• Purpose-built exascale AI storage with GPU-local caching for high-throughput data access

• Serverless RL, the first publicly available fully managed reinforcement learning infrastructure

• Weights and Biases platform integration for RL and agent development workflows

• No egress fees with multi-cloud data portability

• 1.3 gigawatts of contracted power capacity across 32+ data centers globally

• Industry-leading MLPerf benchmark results for AI training and inference workloads

See CoreWeave pricing details →