CoreWeave

Teams that have outgrown a general-purpose cloud are the audience CoreWeave writes for. It calls itself the Essential Cloud for AI, and its platform pairs GPU infrastructure with a software layer for training, tuning and serving models. The company was founded in New Jersey in 2017 and says it has been publicly listed on Nasdaq since March 2025.

GPU clusters from Hopper to Vera Rubin

The infrastructure side lists NVIDIA Hopper, Blackwell and Vera Rubin compute at cluster scale, CPU compute for supporting workloads, and bare metal servers with direct hardware access and no virtualisation layer. Around the compute sit a high-bandwidth network fabric, AI object storage, dedicated VAST storage clusters and a Zero Egress Migration programme that moves customer data in without egress fees. CoreWeave Kubernetes Service provides managed Kubernetes, while Mission Control bundles fleet and node lifecycle controllers, observability and security. Tensorizer, a separate tool, loads models faster through optimized serialization. The company quotes figures such as 96 percent goodput and up to 98 percent effective training time on its own site.

CoreWeave Forge and the AI loop

Forge is the newer software half of the platform. CoreWeave describes every production AI system as a loop of running, observing, curating signal, improving and proving the new version is better, and Forge is built to serve each turn. Under training it lists Weights & Biases Models, Serverless SFT, Serverless RL and a preview of Model Distillation. For inference there is Serverless Inference billed per token on open models and Dedicated Inference on chosen GPUs. Agent work has Agent Lens and Sandboxes, and the tools group adds ARIA, a research agent that analyzes runs, a Registry for models, agents and datasets, and reactive Python notebooks.

Pioneers inventing, advancing and transforming with AI

The about page sorts customers into frontier labs training models, builders of AI-native products, and enterprises or institutions reworking mission-critical functions. Solutions pages cover model training, reinforcement learning, agentic AI, inference and VFX rendering, and by industry they cover federal agencies, media and entertainment, and physical AI, meaning compute for robotics and real-world systems. A Physical AI Field Engineering programme places embedded engineers with customer teams. The home page says nine of ten leading foundation model providers use CoreWeave and carries customer stories from Cohere, IBM and Mistral AI.

Key facts

Company: CoreWeave Founded: 2017, New Jersey Listing: Nasdaq, ticker CRWV, since March 2025 (company statement) Data centres: 50+ active across the U.S. and Europe (company statement) Chief executive: Mike Intrator, co-founder Other co-founders: Brian Venturo, Brannin McBee, Peter Salanki Address on site: 290 W Mt Pleasant Ave, Suite 4100, Livingston, NJ 07039 Platform: CoreWeave Infrastructure and CoreWeave Forge